Merge pull request #1277 from roboflow/feat/patch-rich-label-annotator

Patch RichLabelAnnotator, fix in-place functionality of other annotators
bump version from 0.21.0 to 0.22.0rc1
2024-07-03 13:39:10 +02:00 · 2024-07-03 13:15:49 +02:00 · 2024-07-03 13:05:22 +03:00 · 2024-07-03 13:04:46 +03:00 · 2024-07-03 00:44:18 +00:00 · 2024-07-03 00:43:11 +00:00
138 changed files with 13345 additions and 2490 deletions
@@ -4,12 +4,18 @@ on:
  push:
    branches:
      - develop
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event_name == 'push' && github.ref}}
+  cancel-in-progress: true

 permissions:
  contents: write
  pages: write
  pull-requests: write

+
 jobs:
  deploy:
    runs-on: ubuntu-latest
@@ -23,7 +29,7 @@ jobs:
        with:
          python-version: '3.10'
      - name: 📦 Install mkdocs-material
-        run: pip install "mkdocs-material[all]"
+        run: pip install "mkdocs-material"
      - name: 📦 Install mkdocstrings[python]
        run: pip install "mkdocstrings[python]"
      - name: 📦 Install mkdocs-material[imaging]
@@ -0,0 +1,55 @@
+name: Supervision Release Documentation Workflow 📚
+on:
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event_name == 'push' && github.ref}}
+  cancel-in-progress: true
+
+permissions:
+  contents: write
+  pages: write
+  pull-requests: write
+
+
+jobs:
+  doc-build-deploy:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.10"]
+    steps:
+      - name: 🛎️ Checkout
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          ref: ${{ github.head_ref }}
+
+      - name: 🐍 Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.10'
+      - name: 📦 Install mkdocs-material
+        run: pip install "mkdocs-material"
+      - name: 📦 Install mkdocstrings[python]
+        run: pip install "mkdocstrings[python]"
+      - name: 📦 Install mkdocs-material[imaging]
+        run: pip install "mkdocs-material[imaging]"
+      - name: 📦 Install mike
+        run: pip install "mike"
+      - name: 📦 Install mkdocs-git-revision-date-localized-plugin
+        run: pip install "mkdocs-git-revision-date-localized-plugin"
+      - name: 📦 Install JupyterLab
+        run: pip install jupyterlab
+      - name: 📦 Install mkdocs-jupyter
+        run: pip install mkdocs-jupyter
+      - name: 📦 Install mkdocs-git-committers-plugin-2
+        run: pip install mkdocs-git-committers-plugin-2
+      - name: ⚙️ Configure git for github-actions 👷
+        run: |
+          git config --global user.name "github-actions[bot]"
+          git config --global user.email "41898282+github-actions[bot]@users.noreply.github.com"
+      - name: 🚀 Deploy MkDoc-Material 📚
+        run: |
+            latest_tag=$(git describe --tags `git rev-list --tags --max-count=1`)
+            MKDOCS_GIT_COMMITTERS_APIKEY=${{ secrets.GITHUB_TOKEN }} mike deploy --push --update-aliases $latest_tag latest
@@ -6,22 +6,24 @@ on:
      - '[0-9]+.[0-9]+[0-9]+.[0-9]+b[0-9]'
      - '[0-9]+.[0-9]+[0-9]+.[0-9]+rc[0-9]'

-  # Allows you to run this workflow manually from the Actions tab
  workflow_dispatch:

 jobs:
  build-n-publish:
    name: Build and publish to PyPI
    runs-on: ubuntu-latest
-
+    strategy:
+      matrix:
+        python-version: ["3.10"]
    steps:
-      - name: Checkout source
+      - name: 🛎️ Checkout
        uses: actions/checkout@v4
-
-      - name: 🐍 Set up Python 3.8 environment for build
+        with:
+          ref: ${{ github.head_ref }}
+      - name: 🐍 Set up Python ${{ matrix.python-version }}
        uses: actions/setup-python@v5
        with:
-          python-version: "3.8"
+          python-version: ${{ matrix.python-version }}

      - name: 🏗️ Build source and wheel distributions
        run: |
@@ -4,7 +4,6 @@ on:
    tags:
      - '[0-9]+.[0-9]+[0-9]+.[0-9]'

-  # Allows you to run this workflow manually from the Actions tab
  workflow_dispatch:

 jobs:
@@ -12,7 +11,7 @@ jobs:
    runs-on: ubuntu-latest
    strategy:
      matrix:
-        python-version: [3.8]
+        python-version: ["3.10"]
    steps:
      - name: 🛎️ Checkout
        uses: actions/checkout@v4
@@ -7,7 +7,7 @@ ci:

 repos:
  - repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v4.5.0
+    rev: v4.6.0
    hooks:
      - id: end-of-file-fixer
      - id: trailing-whitespace
@@ -27,9 +27,8 @@ repos:
      - id: mixed-line-ending


-
  -   repo: https://github.com/PyCQA/bandit
-      rev: '1.7.8'
+      rev: '1.7.9'
      hooks:
      -   id: bandit
          args: ["-c", "pyproject.toml"]
@@ -46,7 +45,7 @@ repos:


  -   repo: https://github.com/astral-sh/ruff-pre-commit
-      rev: v0.3.2
+      rev: v0.5.0
      hooks:
      -   id: ruff
          args: [--fix, --exit-non-zero-on-fix]
@@ -64,7 +64,7 @@ To run the pre-commit tool, follow these steps:

 3. Run the command `pre-commit run --all-files`. This will execute the pre-commit hooks configured for this project against the modified files. If any issues are found, the pre-commit tool will provide feedback on how to resolve them. Make the necessary changes and re-run the pre-commit command until all issues are resolved.

-4. You can also install pre-commit as a git hook by execute `pre-commit install`. Every time you made `git commit` pre-commit run automatically for you.
+4. You can also install pre-commit as a git hook by executing `pre-commit install`. Every time you do a `git commit` pre-commit run automatically for you.

 ### Docstrings

@@ -18,17 +18,17 @@
 [![downloads](https://img.shields.io/pypi/dm/supervision)](https://pypistats.org/packages/supervision)
 [![license](https://img.shields.io/pypi/l/supervision)](https://github.com/roboflow/supervision/blob/main/LICENSE.md)
 [![python-version](https://img.shields.io/pypi/pyversions/supervision)](https://badge.fury.io/py/supervision)
-[![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/roboflow/supervision/blob/main/demo.ipynb)
-[![Gradio](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https://huggingface.co/spaces/Roboflow/Annotators)
-[![Discord](https://img.shields.io/discord/1159501506232451173)](https://discord.gg/GbfgXGJ8Bk)
-[![Built with Material for MkDocs](https://img.shields.io/badge/Material_for_MkDocs-526CFE?logo=MaterialForMkDocs&logoColor=white)](https://squidfunk.github.io/mkdocs-material/)
+[![colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/roboflow/supervision/blob/main/demo.ipynb)
+[![gradio](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https://huggingface.co/spaces/Roboflow/Annotators)
+[![discord](https://img.shields.io/discord/1159501506232451173)](https://discord.gg/GbfgXGJ8Bk)
+[![built-with-material-for-mkdocs](https://img.shields.io/badge/Material_for_MkDocs-526CFE?logo=MaterialForMkDocs&logoColor=white)](https://squidfunk.github.io/mkdocs-material/)
 </div>

 ## 👋 hello

 **We write your reusable computer vision tools.** Whether you need to load your dataset from your hard drive, draw detections on an image or video, or count how many detections are in a zone. You can count on us! 🤝

-[![supervision-hackfest](https://github.com/roboflow/supervision/assets/26109316/c05cc954-b9a6-4ed5-9a52-d0b4b619ff65)](https://github.com/orgs/roboflow/projects/10)
+[![supervision-hackfest](https://github.com/roboflow/supervision/assets/26109316/c05cc954-b9a6-4ed5-9a52-d0b4b619ff65)](https://github.com/orgs/roboflow/projects)

 ## 💻 install

@@ -39,7 +39,7 @@ Pip install the supervision package in a
 pip install supervision
 ```

-Read more about desktop, headless, and local installation in our [guide](https://roboflow.github.io/supervision/).
+Read more about conda, mamba, and installing from source in our [guide](https://roboflow.github.io/supervision/).

 ## 🔥 quickstart

@@ -71,23 +71,22 @@ len(detections)
    ```python
    import cv2
    import supervision as sv
-    from inference.models.utils import get_roboflow_model
+    from inference import get_model

    image = cv2.imread(...)
-    model = get_roboflow_model(model_id="yolov8s-640", api_key=<ROBOFLOW API KEY>)
+    model = get_model(model_id="yolov8s-640", api_key=<ROBOFLOW API KEY>)
    result = model.infer(image)[0]
    detections = sv.Detections.from_inference(result)

    len(detections)
-    # 5
-
+    # 5
    ```

 </details>

 ### annotators

-Supervision offers a wide range of highly customizable [annotators](https://supervision.roboflow.com/latest/annotators/), allowing you to compose the perfect visualization for your use case.
+Supervision offers a wide range of highly customizable [annotators](https://supervision.roboflow.com/latest/detection/annotators/), allowing you to compose the perfect visualization for your use case.

 ```python
 import cv2
@@ -107,7 +106,7 @@ https://github.com/roboflow/supervision/assets/26109316/691e219c-0565-4403-9218-

 ### datasets

-Supervision provides a set of [utils](https://supervision.roboflow.com/latest/datasets/) that allow you to load, split, merge, and save datasets in one of the supported formats.
+Supervision provides a set of [utils](https://supervision.roboflow.com/latest/datasets/core/) that allow you to load, split, merge, and save datasets in one of the supported formats.

 ```python
 import supervision as sv
@@ -217,19 +216,23 @@ len(dataset)

 ## 🎬 tutorials

-<p align="left">
-<a href="https://youtu.be/uWP6UjDeZvY" title="Speed Estimation & Vehicle Tracking | Computer Vision | Open Source"><img src="https://github.com/SkalskiP/SkalskiP/assets/26109316/61a444c8-b135-48ce-b979-2a5ab47c5a91" alt="Speed Estimation & Vehicle Tracking | Computer Vision | Open Source" width="300px" align="left" /></a>
-<a href="https://youtu.be/uWP6UjDeZvY" title="Speed Estimation & Vehicle Tracking | Computer Vision | Open Source"><strong>Speed Estimation & Vehicle Tracking | Computer Vision | Open Source</strong></a>
-<div><strong>Created: 11 Jan 2024</strong> | <strong>Updated: 11 Jan 2024</strong></div>
-<br/> Learn how to track and estimate the speed of vehicles using YOLO, ByteTrack, and Roboflow Inference. This comprehensive tutorial covers object detection, multi-object tracking, filtering detections, perspective transformation, speed estimation, visualization improvements, and more.</p>
+Want to learn how to use Supervision? Explore our [how-to guides](https://supervision.roboflow.com/develop/how_to/detect_and_annotate/), [end-to-end examples](https://github.com/roboflow/supervision/tree/develop/examples), and [cookbooks](https://supervision.roboflow.com/develop/cookbooks/)!

 <br/>

 <p align="left">
-<a href="https://youtu.be/4Q3ut7vqD5o" title="Traffic Analysis with YOLOv8 and ByteTrack - Vehicle Detection and Tracking"><img src="https://github.com/roboflow/supervision/assets/26109316/54afdf1c-218c-4451-8f12-627fb85f1682" alt="Traffic Analysis with YOLOv8 and ByteTrack - Vehicle Detection and Tracking" width="300px" align="left" /></a>
-<a href="https://youtu.be/4Q3ut7vqD5o" title="Traffic Analysis with YOLOv8 and ByteTrack - Vehicle Detection and Tracking"><strong>Traffic Analysis with YOLOv8 and ByteTrack - Vehicle Detection and Tracking</strong></a>
-<div><strong>Created: 6 Sep 2023</strong> | <strong>Updated: 6 Sep 2023</strong></div>
-<br/> In this video, we explore real-time traffic analysis using YOLOv8 and ByteTrack to detect and track vehicles on aerial images. Harnessing the power of Python and Supervision, we delve deep into assigning cars to specific entry zones and understanding their direction of movement. By visualizing their paths, we gain insights into traffic flow across bustling roundabouts... </p>
+<a href="https://youtu.be/hAWpsIuem10" title="Dwell Time Analysis with Computer Vision | Real-Time Stream Processing"><img src="https://github.com/SkalskiP/SkalskiP/assets/26109316/a742823d-c158-407d-b30f-063a5d11b4e1" alt="Dwell Time Analysis with Computer Vision | Real-Time Stream Processing" width="300px" align="left" /></a>
+<a href="https://youtu.be/hAWpsIuem10" title="Dwell Time Analysis with Computer Vision | Real-Time Stream Processing"><strong>Dwell Time Analysis with Computer Vision | Real-Time Stream Processing</strong></a>
+<div><strong>Created: 5 Apr 2024</strong></div>
+<br/>Learn how to use computer vision to analyze wait times and optimize processes. This tutorial covers object detection, tracking, and calculating time spent in designated zones. Use these techniques to improve customer experience in retail, traffic management, or other scenarios.</p>
+
+<br/>
+
+<p align="left">
+<a href="https://youtu.be/uWP6UjDeZvY" title="Speed Estimation & Vehicle Tracking | Computer Vision | Open Source"><img src="https://github.com/SkalskiP/SkalskiP/assets/26109316/61a444c8-b135-48ce-b979-2a5ab47c5a91" alt="Speed Estimation & Vehicle Tracking | Computer Vision | Open Source" width="300px" align="left" /></a>
+<a href="https://youtu.be/uWP6UjDeZvY" title="Speed Estimation & Vehicle Tracking | Computer Vision | Open Source"><strong>Speed Estimation & Vehicle Tracking | Computer Vision | Open Source</strong></a>
+<div><strong>Created: 11 Jan 2024</strong></div>
+<br/>Learn how to track and estimate the speed of vehicles using YOLO, ByteTrack, and Roboflow Inference. This comprehensive tutorial covers object detection, multi-object tracking, filtering detections, perspective transformation, speed estimation, visualization improvements, and more.</p>

 ## 💜 built with supervision

@@ -19,13 +19,13 @@ as an extra within the Supervision package.
    ```

 <div class="md-typeset">
-  <h2>download_assets</h2>
+    <h2><a href="#supervision.assets.downloader.download_assets.download_assets">download_assets</a></h2>
 </div>

 :::supervision.assets.downloader.download_assets

 <div class="md-typeset">
-  <h2>VideoAssets</h2>
+    <h2><a href="#supervision.assets.downloader.download_assets.VideoAssets">VideoAssets</a></h2>
 </div>

 :::supervision.assets.list.VideoAssets
@@ -1,3 +1,148 @@
+### 0.21.0 <small>Jun 5, 2024</small>
+
+- Added [#500](https://github.com/roboflow/supervision/pull/500): [`sv.Detections.with_nmm`](https://supervision.roboflow.com/develop/detection/core/#supervision.detection.core.Detections.with_nmm) to perform non-maximum merging on the current set of object detections.
+
+- Added [#1221](https://github.com/roboflow/supervision/pull/1221): [`sv.Detections.from_lmm`](https://supervision.roboflow.com/develop/detection/core/#supervision.detection.core.Detections.from_lmm) allowing to parse Large Multimodal Model (LMM) text result into [`sv.Detections`](https://supervision.roboflow.com/develop/detection/core/) object. For now `from_lmm` supports only [PaliGemma](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/how-to-finetune-paligemma-on-detection-dataset.ipynb) result parsing.
+
+```python
+import supervision as sv
+
+paligemma_result = "<loc0256><loc0256><loc0768><loc0768> cat"
+detections = sv.Detections.from_lmm(
+    sv.LMM.PALIGEMMA,
+    paligemma_result,
+    resolution_wh=(1000, 1000),
+    classes=['cat', 'dog']
+)
+detections.xyxy
+# array([[250., 250., 750., 750.]])
+
+detections.class_id
+# array([0])
+```
+
+- Added [#1236](https://github.com/roboflow/supervision/pull/1236): [`sv.VertexLabelAnnotator`](https://supervision.roboflow.com/develop/keypoint/annotators/#supervision.keypoint.annotators.EdgeAnnotator.annotate) allowing to annotate every vertex of a keypoint skeleton with custom text and color.
+
+```python
+import supervision as sv
+
+image = ...
+key_points = sv.KeyPoints(...)
+
+edge_annotator = sv.EdgeAnnotator(
+    color=sv.Color.GREEN,
+    thickness=5
+)
+annotated_frame = edge_annotator.annotate(
+    scene=image.copy(),
+    key_points=key_points
+)
+```
+
+- Added [#1147](https://github.com/roboflow/supervision/pull/1147): [`sv.KeyPoints.from_inference`](https://supervision.roboflow.com/develop/keypoint/core/#supervision.keypoint.core.KeyPoints.from_inference) allowing to create [`sv.KeyPoints`](https://supervision.roboflow.com/develop/keypoint/core/#supervision.keypoint.core.KeyPoints) from [Inference](https://github.com/roboflow/inference) result.
+
+- Added [#1138](https://github.com/roboflow/supervision/pull/1138): [`sv.KeyPoints.from_yolo_nas`](https://supervision.roboflow.com/develop/keypoint/core/#supervision.keypoint.core.KeyPoints.from_yolo_nas) allowing to create [`sv.KeyPoints`](https://supervision.roboflow.com/develop/keypoint/core/#supervision.keypoint.core.KeyPoints) from [YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md) result.
+
+- Added [#1163](https://github.com/roboflow/supervision/pull/1163): [`sv.mask_to_rle`](https://supervision.roboflow.com/develop/datasets/utils/#supervision.dataset.utils.rle_to_mask) and [`sv.rle_to_mask`](https://supervision.roboflow.com/develop/datasets/utils/#supervision.dataset.utils.rle_to_mask) allowing for easy conversion between mask and rle formats.
+
+- Changed [#1236](https://github.com/roboflow/supervision/pull/1236): [`sv.InferenceSlicer`](https://supervision.roboflow.com/develop/detection/tools/inference_slicer/) allowing to select overlap filtering strategy (`NONE`, `NON_MAX_SUPPRESSION` and `NON_MAX_MERGE`).
+
+- Changed [#1178](https://github.com/roboflow/supervision/pull/1178): [`sv.InferenceSlicer`](https://supervision.roboflow.com/develop/detection/tools/inference_slicer/) adding instance segmentation model support.
+
+```python
+import cv2
+import numpy as np
+import supervision as sv
+from inference import get_model
+
+model = get_model(model_id="yolov8x-seg-640")
+image = cv2.imread(<SOURCE_IMAGE_PATH>)
+
+def callback(image_slice: np.ndarray) -> sv.Detections:
+    results = model.infer(image_slice)[0]
+    return sv.Detections.from_inference(results)
+
+slicer = sv.InferenceSlicer(callback = callback)
+detections = slicer(image)
+
+mask_annotator = sv.MaskAnnotator()
+label_annotator = sv.LabelAnnotator()
+
+annotated_image = mask_annotator.annotate(
+    scene=image, detections=detections)
+annotated_image = label_annotator.annotate(
+    scene=annotated_image, detections=detections)
+```
+
+- Changed [#1228](https://github.com/roboflow/supervision/pull/1228): [`sv.LineZone`](https://supervision.roboflow.com/develop/detection/tools/line_zone/) making it 10-20 times faster, depending on the use case.
+
+- Changed [#1163](https://github.com/roboflow/supervision/pull/1163): [`sv.DetectionDataset.from_coco`](https://supervision.roboflow.com/develop/datasets/core/#supervision.dataset.core.DetectionDataset.from_coco) and [`sv.DetectionDataset.as_coco`](https://supervision.roboflow.com/develop/datasets/core/#supervision.dataset.core.DetectionDataset.as_coco) adding support for run-length encoding (RLE) mask format.
+
+### 0.20.0 <small>April 24, 2024</small>
+
+- Added [#1128](https://github.com/roboflow/supervision/pull/1128): [`sv.KeyPoints`](/0.20.0/keypoint/core/#supervision.keypoint.core.KeyPoints) to provide initial support for pose estimation and broader keypoint detection models.
+
+- Added [#1128](https://github.com/roboflow/supervision/pull/1128): [`sv.EdgeAnnotator`](/0.20.0/keypoint/annotators/#supervision.keypoint.annotators.EdgeAnnotator) and [`sv.VertexAnnotator`](/0.20.0/keypoint/annotators/#supervision.keypoint.annotators.VertexAnnotator) to enable rendering of results from keypoint detection models.
+
+```python
+import cv2
+import supervision as sv
+from ultralytics import YOLO
+
+image = cv2.imread(<SOURCE_IMAGE_PATH>)
+model = YOLO('yolov8l-pose')
+
+result = model(image, verbose=False)[0]
+keypoints = sv.KeyPoints.from_ultralytics(result)
+
+edge_annotators = sv.EdgeAnnotator(color=sv.Color.GREEN, thickness=5)
+annotated_image = edge_annotators.annotate(image.copy(), keypoints)
+```
+
+- Changed [#1037](https://github.com/roboflow/supervision/pull/1037): [`sv.LabelAnnotator`](/0.20.0/annotators/#supervision.annotators.core.LabelAnnotator) by adding an additional `corner_radius` argument that allows for rounding the corners of the bounding box.
+
+- Changed [#1109](https://github.com/roboflow/supervision/pull/1109): [`sv.PolygonZone`](/0.20.0/detection/tools/polygon_zone/#supervision.detection.tools.polygon_zone.PolygonZone) such that the `frame_resolution_wh` argument is no longer required to initialize `sv.PolygonZone`.
+
+!!! failure "Deprecated"
+
+    The `frame_resolution_wh` parameter in `sv.PolygonZone` is deprecated and will be removed in `supervision-0.24.0`.
+
+- Changed [#1084](https://github.com/roboflow/supervision/pull/1084): [`sv.get_polygon_center`](/0.20.0/utils/geometry/#supervision.geometry.core.utils.get_polygon_center) to calculate a more accurate polygon centroid.
+
+- Changed [#1069](https://github.com/roboflow/supervision/pull/1069): [`sv.Detections.from_transformers`](/0.20.0/detection/core/#supervision.detection.core.Detections.from_transformers) by adding support for Transformers segmentation models and extract class names values.
+
+```python
+import torch
+import supervision as sv
+from PIL import Image
+from transformers import DetrImageProcessor, DetrForSegmentation
+
+processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50-panoptic")
+model = DetrForSegmentation.from_pretrained("facebook/detr-resnet-50-panoptic")
+
+image = Image.open(<SOURCE_IMAGE_PATH>)
+inputs = processor(images=image, return_tensors="pt")
+
+with torch.no_grad():
+    outputs = model(**inputs)
+
+width, height = image.size
+target_size = torch.tensor([[height, width]])
+results = processor.post_process_segmentation(
+    outputs=outputs, target_sizes=target_size)[0]
+detections = sv.Detections.from_transformers(results, id2label=model.config.id2label)
+
+mask_annotator = sv.MaskAnnotator()
+label_annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER)
+
+annotated_image = mask_annotator.annotate(
+    scene=image, detections=detections)
+annotated_image = label_annotator.annotate(
+    scene=annotated_image, detections=detections)
+```
+
+- Fixed [#787](https://github.com/roboflow/supervision/pull/787): [`sv.ByteTrack.update_with_detections`](/0.20.0/trackers/#supervision.tracker.byte_tracker.core.ByteTrack.update_with_detections) which was removing segmentation masks while tracking. Now, `ByteTrack` can be used alongside segmentation models.
+
 ### 0.19.0 <small>March 15, 2024</small>

 - Added [#818](https://github.com/roboflow/supervision/pull/818): [`sv.CSVSink`](/0.19.0/detection/tools/save_detections/#supervision.detection.tools.csv_sink.CSVSink) allowing for the straightforward saving of image, video, or stream inference results in a `.csv` file.
@@ -1,7 +1,6 @@
 ---
 template: cookbooks.html
 comments: true
-status: new
 hide:
  - navigation
  - toc
@@ -1,5 +1,6 @@
 ---
 comments: true
+status: new
 ---

 # Datasets
@@ -0,0 +1,18 @@
+---
+comments: true
+status: new
+---
+
+# Datasets Utils
+
+<div class="md-typeset">
+  <h2><a href="#supervision.dataset.utils.rle_to_mask">rle_to_mask</a></h2>
+</div>
+
+:::supervision.dataset.utils.rle_to_mask
+
+<div class="md-typeset">
+  <h2><a href="#supervision.dataset.utils.mask_to_rle">mask_to_rle</a></h2>
+</div>
+
+:::supervision.dataset.utils.mask_to_rle
@@ -12,7 +12,8 @@ These features are phased out due to better alternatives or potential issues in
 - The method `Color.green()` is deprecated and will be removed in `supervision-0.22.0`. Use the constant `Color.GREEN` instead.
 - The method `Color.blue()` is deprecated and will be removed in `supervision-0.22.0`. Use the constant `Color.BLUE` instead.
 - The method [`ColorPalette.default()`](draw/color.md/#supervision.draw.color.ColorPalette.default) is deprecated and will be removed in `supervision-0.22.0`. Use the constant [`ColorPalette.DEFAULT`](draw/color.md/#supervision.draw.color.ColorPalette.DEFAULT) instead.
- `BoxAnnotator` is deprecated and will be removed in `supervision-0.22.0`. Use [`BoundingBoxAnnotator`](annotators.md/#supervision.annotators.core.BoundingBoxAnnotator) and [`LabelAnnotator`](annotators.md/#supervision.annotators.core.LabelAnnotator) instead.
+- `BoxAnnotator` is deprecated and will be removed in `supervision-0.22.0`. Use [`BoundingBoxAnnotator`](detection/annotators.md/#supervision.annotators.core.BoundingBoxAnnotator) and [`LabelAnnotator`](detection/annotators.md/#supervision.annotators.core.LabelAnnotator) instead.
 - The method [`FPSMonitor.__call__`](utils/video.md/#supervision.utils.video.FPSMonitor.__call__) is deprecated and will be removed in `supervision-0.22.0`. Use the attribute [`FPSMonitor.fps`](utils/video.md/#supervision.utils.video.FPSMonitor.fps) instead.
 - The `track_buffer`, `track_thresh`, and `match_thresh` parameters in [`ByterTrack`](trackers.md/#supervision.tracker.byte_tracker.core.ByteTrack) are deprecated and will be removed in `supervision-0.23.0`. Use `lost_track_buffer,` `track_activation_threshold`, and `minimum_matching_threshold` instead.
 - The `triggering_position ` parameter in [`sv.PolygonZone`](detection/tools/polygon_zone.md/#supervision.detection.tools.polygon_zone.PolygonZone) is deprecated and will be removed in `supervision-0.23.0`. Use `triggering_anchors ` instead.
+- The `frame_resolution_wh ` parameter in [`sv.PolygonZone`](detection/tools/polygon_zone.md/#supervision.detection.tools.polygon_zone.PolygonZone) is deprecated and will be removed in `supervision-0.24.0`.
@@ -1,6 +1,5 @@
 ---
 comments: true
-status: new
 ---

 # Annotators
@@ -260,15 +259,53 @@ status: new
 === "Label"

    ```python
-     import supervision as sv
+    import supervision as sv

    image = ...
    detections = sv.Detections(...)

+    labels = [
+        f"{class_name} {confidence:.2f}"
+        for class_name, confidence
+        in zip(detections['class_name'], detections.confidence)
+    ]
+
    label_annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER)
    annotated_frame = label_annotator.annotate(
        scene=image.copy(),
-        detections=detections
+        detections=detections,
+        labels=labels
+    )
+    ```
+
+    <div class="result" markdown>
+
+    ![label-annotator-example](https://media.roboflow.com/supervision-annotator-examples/label-annotator-example-purple.png){ align=center width="800" }
+
+    </div>
+
+=== "RichLabel"
+
+    ```python
+    import supervision as sv
+
+    image = ...
+    detections = sv.Detections(...)
+
+    labels = [
+        f"{class_name} {confidence:.2f}"
+        for class_name, confidence
+        in zip(detections['class_name'], detections.confidence)
+    ]
+
+    rich_label_annotator = sv.RichLabelAnnotator(
+        font_path=".../font.ttf",
+        text_position=sv.Position.CENTER
+    )
+    annotated_frame = label_annotator.annotate(
+        scene=image.copy(),
+        detections=detections,
+        labels=labels
    )
    ```

@@ -281,7 +318,7 @@ status: new
 === "Crop"

    ```python
-     import supervision as sv
+    import supervision as sv

    image = ...
    detections = sv.Detections(...)
@@ -485,6 +522,12 @@ status: new

 :::supervision.annotators.core.LabelAnnotator

+<div class="md-typeset">
+    <h2><a href="#supervision.annotators.core.RichLabelAnnotator">RichLabelAnnotator</a></h2>
+</div>
+
+:::supervision.annotators.core.RichLabelAnnotator
+
 <div class="md-typeset">
    <h2><a href="#supervision.annotators.core.BlurAnnotator">BlurAnnotator</a></h2>
 </div>
@@ -0,0 +1,30 @@
+---
+comments: true
+status: new
+---
+
+# Double Detection Filter
+
+<div class="md-typeset">
+  <h2><a href="#supervision.detection.overlap_filter.OverlapFilter">OverlapFilter</a></h2>
+</div>
+
+:::supervision.detection.overlap_filter.OverlapFilter
+
+<div class="md-typeset">
+  <h2><a href="#supervision.detection.overlap_filter.box_non_max_suppression">box_non_max_suppression</a></h2>
+</div>
+
+:::supervision.detection.overlap_filter.box_non_max_suppression
+
+<div class="md-typeset">
+  <h2><a href="#supervision.detection.overlap_filter.mask_non_max_suppression">mask_non_max_suppression</a></h2>
+</div>
+
+:::supervision.detection.overlap_filter.mask_non_max_suppression
+
+<div class="md-typeset">
+  <h2><a href="#supervision.detection.overlap_filter.box_non_max_merge">box_non_max_merge</a></h2>
+</div>
+
+:::supervision.detection.overlap_filter.box_non_max_merge
@@ -0,0 +1,17 @@
+---
+comments: true
+---
+
+# Metrics
+
+<div class="md-typeset">
+    <h2><a href="#supervision.metrics.detection.ConfusionMatrix">ConfusionMatrix</a></h2>
+</div>
+
+:::supervision.metrics.detection.ConfusionMatrix
+
+<div class="md-typeset">
+    <h2><a href="#supervision.metrics.detection.MeanAveragePrecision">MeanAveragePrecision</a></h2>
+</div>
+
+:::supervision.metrics.detection.MeanAveragePrecision
@@ -1,5 +1,6 @@
 ---
 comments: true
+status: new
 ---

 # InferenceSlicer
@@ -1,5 +1,6 @@
 ---
 comments: true
+status: new
 ---

 <div class="md-typeset">
@@ -1,6 +1,5 @@
 ---
 comments: true
-status: new
 ---

 # Save Detections
@@ -1,72 +1,89 @@
 ---
 comments: true
-status: new
 ---

 # Detection Utils

 <div class="md-typeset">
-  <h2>box_iou_batch</h2>
+  <h2><a href="#supervision.detection.utils.box_iou_batch">box_iou_batch</a></h2>
 </div>

 :::supervision.detection.utils.box_iou_batch

 <div class="md-typeset">
-  <h2>mask_iou_batch</h2>
+  <h2><a href="#supervision.detection.utils.mask_iou_batch">mask_iou_batch</a></h2>
 </div>

 :::supervision.detection.utils.mask_iou_batch

 <div class="md-typeset">
-  <h2>box_non_max_suppression</h2>
-</div>
-
-:::supervision.detection.utils.box_non_max_suppression
-
-<div class="md-typeset">
-  <h2>mask_non_max_suppression</h2>
-</div>
-
-:::supervision.detection.utils.mask_non_max_suppression
-
-<div class="md-typeset">
-    <h2>polygon_to_mask</h2>
+  <h2><a href="#supervision.detection.utils.polygon_to_mask">polygon_to_mask</a></h2>
 </div>

 :::supervision.detection.utils.polygon_to_mask

 <div class="md-typeset">
-    <h2>mask_to_xyxy</h2>
+  <h2><a href="#supervision.detection.utils.mask_to_xyxy">mask_to_xyxy</a></h2>
 </div>

 :::supervision.detection.utils.mask_to_xyxy

 <div class="md-typeset">
-    <h2>mask_to_polygons</h2>
+  <h2><a href="#supervision.detection.utils.mask_to_polygons">mask_to_polygons</a></h2>
 </div>

 :::supervision.detection.utils.mask_to_polygons

 <div class="md-typeset">
-    <h2>polygon_to_xyxy</h2>
+  <h2><a href="#supervision.detection.utils.polygon_to_xyxy">polygon_to_xyxy</a></h2>
 </div>

 :::supervision.detection.utils.polygon_to_xyxy

 <div class="md-typeset">
-    <h2>filter_polygons_by_area</h2>
+  <h2><a href="#supervision.detection.utils.filter_polygons_by_area">filter_polygons_by_area</a></h2>
 </div>

 :::supervision.detection.utils.filter_polygons_by_area

 <div class="md-typeset">
-    <h2>move_boxes</h2>
+  <h2><a href="#supervision.detection.utils.move_boxes">move_boxes</a></h2>
 </div>

 :::supervision.detection.utils.move_boxes

 <div class="md-typeset">
-    <h2>scale_boxes</h2>
+  <h2><a href="#supervision.detection.utils.move_masks">move_masks</a></h2>
+</div>
+
+:::supervision.detection.utils.move_masks
+
+<div class="md-typeset">
+  <h2><a href="#supervision.detection.utils.scale_boxes">scale_boxes</a></h2>
 </div>

 :::supervision.detection.utils.scale_boxes
+
+<div class="md-typeset">
+  <h2><a href="#supervision.detection.utils.clip_boxes">clip_boxes</a></h2>
+</div>
+
+:::supervision.detection.utils.clip_boxes
+
+<div class="md-typeset">
+  <h2><a href="#supervision.detection.utils.pad_boxes">pad_boxes</a></h2>
+</div>
+
+:::supervision.detection.utils.pad_boxes
+
+<div class="md-typeset">
+  <h2><a href="#supervision.detection.utils.contains_holes">contains_holes</a></h2>
+</div>
+
+:::supervision.detection.utils.contains_holes
+
+<div class="md-typeset">
+  <h2><a href="#supervision.detection.utils.contains_multiple_segments">contains_multiple_segments</a></h2>
+</div>
+
+:::supervision.detection.utils.contains_multiple_segments
@@ -1,13 +0,0 @@
---
-comments: true
---
-
-# Color
-
-:::supervision.draw.color.Color
-
-<div class="md-typeset">
-  <h2>ColorPalette</h2>
-</div>
-
-:::supervision.draw.color.ColorPalette
@@ -1,53 +0,0 @@
---
-comments: true
---
-
-# Draw Utils
-
-<div class="md-typeset">
-  <h2>draw_line</h2>
-</div>
-
-:::supervision.draw.utils.draw_line
-
-<div class="md-typeset">
-  <h2>draw_rectangle</h2>
-</div>
-
-:::supervision.draw.utils.draw_rectangle
-
-<div class="md-typeset">
-  <h2>draw_filled_rectangle</h2>
-</div>
-
-:::supervision.draw.utils.draw_filled_rectangle
-
-<div class="md-typeset">
-  <h2>draw_polygon</h2>
-</div>
-
-:::supervision.draw.utils.draw_polygon
-
-<div class="md-typeset">
-  <h2>draw_text</h2>
-</div>
-
-:::supervision.draw.utils.draw_text
-
-<div class="md-typeset">
-  <h2>draw_image</h2>
-</div>
-
-:::supervision.draw.utils.draw_image
-
-<div class="md-typeset">
-  <h2>calculate_dynamic_font_scale</h2>
-</div>
-
-:::supervision.draw.utils.calculate_dynamic_text_scale
-
-<div class="md-typeset">
-  <h2>calculate_dynamic_line_thickness</h2>
-</div>
-
-:::supervision.draw.utils.calculate_dynamic_line_thickness
@@ -1,7 +0,0 @@
---
-comments: true
---
-
-# Position
-
-:::supervision.geometry.core.Position
@@ -4,17 +4,31 @@ comments: true

 # Detect and Annotate

-Supervision offers a streamlined solution to effortlessly annotate predictions from a
-range of object detection and segmentation models. This guide demonstrates how to
-execute inference using the YOLOv8 model with either the
-[Inference](https://github.com/roboflow/inference) or
-[Ultralytics](https://github.com/ultralytics/ultralytics) packages. Following this,
-you'll learn how to import these predictions into Supervision for image annotation
-purposes.
+Supervision provides a seamless process for annotating predictions generated by various
+object detection and segmentation models. This guide shows how to perform inference
+with the  [Inference](https://github.com/roboflow/inference),
+[Ultralytics](https://github.com/ultralytics/ultralytics) or
+[Transformers](https://github.com/huggingface/transformers) packages. Following this,
+you'll learn how to import these predictions into Supervision and use them to annotate
+source image.

-## Run Inference
+![basic-annotation](https://media.roboflow.com/supervision_detect_and_annotate_example_1.png)

-First, you'll need to obtain predictions from your object detection or segmentation model.
+## Run Detection
+
+First, you'll need to obtain predictions from your object detection or segmentation
+model.
+
+=== "Inference"
+
+    ```python
+    import cv2
+    from inference import get_model
+
+    model = get_model(model_id="yolov8n-640")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model.infer(image)[0]
+    ```

 === "Ultralytics"

@@ -23,105 +37,115 @@ First, you'll need to obtain predictions from your object detection or segmentat
    from ultralytics import YOLO

    model = YOLO("yolov8n.pt")
-    image = cv2.imread(<PATH TO IMAGE>)
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
    results = model(image)[0]
    ```

-=== "Inference"
+=== "Transformers"

    ```python
-    import cv2
-    from inference.models.utils import get_roboflow_model
+    import torch
+    from PIL import Image
+    from transformers import DetrImageProcessor, DetrForObjectDetection

-    model = get_roboflow_model(model_id="yolov8n-640", api_key=<ROBOFLOW API KEY>)
-    image = cv2.imread(<PATH TO IMAGE>)
-    results = model.infer(image)[0]
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+
+    image = Image.open(<SOURCE_IMAGE_PATH>)
+    inputs = processor(images=image, return_tensors="pt")
+
+    with torch.no_grad():
+        outputs = model(**inputs)
+
+    width, height = image.size
+    target_size = torch.tensor([[height, width]])
+    results = processor.post_process_object_detection(
+        outputs=outputs, target_sizes=target_size)[0]
    ```

 ## Load Predictions into Supervision

 Now that we have predictions from a model, we can load them into Supervision.

-=== "Ultralytics"
-
-    We can do so using the [`sv.Detections.from_ultralytics`](detection/core/#supervision.detection.core.Detections.from_ultralytics) method, which accepts model results from both detection and segmentation models.
-
-    ```python
-    import cv2
-    import supervision as sv
-    from ultralytics import YOLO
-
-    model = YOLO("yolov8n.pt")
-    image = cv2.imread(<PATH TO IMAGE>)
-    results = model(image)[0]
-    detections = sv.Detections.from_ultralytics(results)
-    ```
-
 === "Inference"

    We can do so using the [`sv.Detections.from_inference`](detection/core/#supervision.detection.core.Detections.from_inference) method, which accepts model results from both detection and segmentation models.

-    ```python
+    ```{ .py hl_lines="2 8" }
    import cv2
    import supervision as sv
-    from inference.models.utils import get_roboflow_model
+    from inference import get_model

-    model = get_roboflow_model(model_id="yolov8n-640", api_key=<ROBOFLOW API KEY>
-    image = cv2.imread(<PATH TO IMAGE>)
+    model = get_model(model_id="yolov8n-640")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
    results = model.infer(image)[0]
    detections = sv.Detections.from_inference(results)
    ```

-You can conveniently load predictions from other computer vision frameworks and libraries using:
-
- [`from_deepsparse`](detection/core/#supervision.detection.core.Detections.from_deepsparse) ([Deepsparse](https://github.com/neuralmagic/deepsparse))
- [`from_detectron2`](detection/core/#supervision.detection.core.Detections.from_detectron2) ([Detectron2](https://github.com/facebookresearch/detectron2))
- [`from_mmdetection`](detection/core/#supervision.detection.core.Detections.from_mmdetection) ([MMDetection](https://github.com/open-mmlab/mmdetection))
- [`from_inference`](detection/core/#supervision.detection.core.Detections.from_inference) ([Roboflow Inference](https://github.com/roboflow/inference))
- [`from_sam`](detection/core/#supervision.detection.core.Detections.from_sam) ([Segment Anything Model](https://github.com/facebookresearch/segment-anything))
- [`from_transformers`](detection/core/#supervision.detection.core.Detections.from_transformers) ([HuggingFace Transformers](https://github.com/huggingface/transformers))
- [`from_yolo_nas`](detection/core/#supervision.detection.core.Detections.from_yolo_nas) ([YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md))
-
-## Annotate Image
-
-Finally, we can annotate the image with the predictions. Since we are working with an object detection model, we will use the [`sv.BoundingBoxAnnotator`](annotators/#supervision.annotators.core.BoundingBoxAnnotator) and [`sv.LabelAnnotator`](annotators/#supervision.annotators.core.LabelAnnotator) classes. If you are running the segmentation model [`sv.MaskAnnotator`](annotators/#supervision.annotators.core.MaskAnnotator) is a drop-in replacement for [`sv.BoundingBoxAnnotator`](annotators/#supervision.annotators.core.BoundingBoxAnnotator) that will allow you to draw masks instead of boxes.
-
 === "Ultralytics"

-    ```python
+    We can do so using the [`sv.Detections.from_ultralytics`](detection/core/#supervision.detection.core.Detections.from_ultralytics) method, which accepts model results from both detection and segmentation models.
+
+    ```{ .py hl_lines="2 8" }
    import cv2
    import supervision as sv
    from ultralytics import YOLO

    model = YOLO("yolov8n.pt")
-    image = cv2.imread(<PATH TO IMAGE>)
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
    results = model(image)[0]
    detections = sv.Detections.from_ultralytics(results)
-
-    bounding_box_annotator = sv.BoundingBoxAnnotator()
-    label_annotator = sv.LabelAnnotator()
-
-    labels = [
-        model.model.names[class_id]
-        for class_id
-        in detections.class_id
-    ]
-
-    annotated_image = bounding_box_annotator.annotate(
-        scene=image, detections=detections)
-    annotated_image = label_annotator.annotate(
-        scene=annotated_image, detections=detections, labels=labels)
    ```

+=== "Transformers"
+
+    We can do so using the [`sv.Detections.from_transformers`](detection/core/#supervision.detection.core.Detections.from_transformers) method, which accepts model results from both detection and segmentation models.
+
+    ```{ .py hl_lines="2 19-21" }
+    import torch
+    import supervision as sv
+    from PIL import Image
+    from transformers import DetrImageProcessor, DetrForObjectDetection
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+
+    image = Image.open(<SOURCE_IMAGE_PATH>)
+    inputs = processor(images=image, return_tensors="pt")
+
+    with torch.no_grad():
+        outputs = model(**inputs)
+
+    width, height = image.size
+    target_size = torch.tensor([[height, width]])
+    results = processor.post_process_object_detection(
+        outputs=outputs, target_sizes=target_size)[0]
+    detections = sv.Detections.from_transformers(
+        transformers_results=results,
+        id2label=model.config.id2label)
+    ```
+
+You can load predictions from other computer vision frameworks and libraries using:
+
+- [`from_deepsparse`](/latest/detection/core/#supervision.detection.core.Detections.from_deepsparse) ([Deepsparse](https://github.com/neuralmagic/deepsparse))
+- [`from_detectron2`](/latest/detection/core/#supervision.detection.core.Detections.from_detectron2) ([Detectron2](https://github.com/facebookresearch/detectron2))
+- [`from_mmdetection`](/latest/detection/core/#supervision.detection.core.Detections.from_mmdetection) ([MMDetection](https://github.com/open-mmlab/mmdetection))
+- [`from_sam`](/latest/detection/core/#supervision.detection.core.Detections.from_sam) ([Segment Anything Model](https://github.com/facebookresearch/segment-anything))
+- [`from_yolo_nas`](/latest/detection/core/#supervision.detection.core.Detections.from_yolo_nas) ([YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md))
+
+## Annotate Image with Detections
+
+Finally, we can annotate the image with the predictions. Since we are working with an object detection model, we will use the [`sv.BoundingBoxAnnotator`](/latest/annotators/#supervision.annotators.core.BoundingBoxAnnotator) and [`sv.LabelAnnotator`](/latest/annotators/#supervision.annotators.core.LabelAnnotator) classes.
+
 === "Inference"

-    ```python
+    ```{ .py hl_lines="10-16" }
    import cv2
    import supervision as sv
-    from inference.models.utils import get_roboflow_model
+    from inference import get_model

-    model = get_roboflow_model(model_id="yolov8n-640", api_key=<ROBOFLOW API KEY>
-    image = cv2.imread(<PATH TO IMAGE>)
+    model = get_model(model_id="yolov8n-640")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
    results = model.infer(image)[0]
    detections = sv.Detections.from_inference(results)

@@ -134,12 +158,253 @@ Finally, we can annotate the image with the predictions. Since we are working wi
        scene=annotated_image, detections=detections)
    ```

-![Predictions plotted on an image](https://media.roboflow.com/supervision_annotate_example.png)
+=== "Ultralytics"

-## Display Annotated Image
+    ```{ .py hl_lines="10-16" }
+    import cv2
+    import supervision as sv
+    from ultralytics import YOLO

-To display the annotated image in Jupyter Notebook or Google Colab, use the [`sv.plot_image`](utils/notebook/#supervision.utils.notebook.plot_image) function.
+    model = YOLO("yolov8n.pt")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model(image)[0]
+    detections = sv.Detections.from_ultralytics(results)

-```python
-sv.plot_image(annotated_image)
-```
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+=== "Transformers"
+
+    ```{ .py hl_lines="23-30" }
+    import torch
+    import supervision as sv
+    from PIL import Image
+    from transformers import DetrImageProcessor, DetrForObjectDetection
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+
+    image = Image.open(<SOURCE_IMAGE_PATH>)
+    inputs = processor(images=image, return_tensors="pt")
+
+    with torch.no_grad():
+        outputs = model(**inputs)
+
+    width, height = image.size
+    target_size = torch.tensor([[height, width]])
+    results = processor.post_process_object_detection(
+        outputs=outputs, target_sizes=target_size)[0]
+    detections = sv.Detections.from_transformers(
+        transformers_results=results,
+        id2label=model.config.id2label)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+![basic-annotation](https://media.roboflow.com/supervision_detect_and_annotate_example_1.png)
+
+## Display Custom Labels
+
+By default, [`sv.LabelAnnotator`](/latest/annotators/#supervision.annotators.core.LabelAnnotator)
+will label each detection with its `class_name` (if possible) or `class_id`. You can
+override this behavior by passing a list of custom `labels` to the `annotate` method.
+
+=== "Inference"
+
+    ```{ .py hl_lines="13-17 22" }
+    import cv2
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8n-640")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model.infer(image)[0]
+    detections = sv.Detections.from_inference(results)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    labels = [
+        f"{class_name} {confidence:.2f}"
+        for class_name, confidence
+        in zip(detections['class_name'], detections.confidence)
+    ]
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections, labels=labels)
+    ```
+
+=== "Ultralytics"
+
+    ```{ .py hl_lines="13-17 22" }
+    import cv2
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8n.pt")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model(image)[0]
+    detections = sv.Detections.from_ultralytics(results)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    labels = [
+        f"{class_name} {confidence:.2f}"
+        for class_name, confidence
+        in zip(detections['class_name'], detections.confidence)
+    ]
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections, labels=labels)
+    ```
+
+=== "Transformers"
+
+    ```{ .py hl_lines="26-30 35" }
+    import torch
+    import supervision as sv
+    from PIL import Image
+    from transformers import DetrImageProcessor, DetrForObjectDetection
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+
+    image = Image.open(<SOURCE_IMAGE_PATH>)
+    inputs = processor(images=image, return_tensors="pt")
+
+    with torch.no_grad():
+        outputs = model(**inputs)
+
+    width, height = image.size
+    target_size = torch.tensor([[height, width]])
+    results = processor.post_process_object_detection(
+        outputs=outputs, target_sizes=target_size)[0]
+    detections = sv.Detections.from_transformers(
+        transformers_results=results,
+        id2label=model.config.id2label)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    labels = [
+        f"{class_name} {confidence:.2f}"
+        for class_name, confidence
+        in zip(detections['class_name'], detections.confidence)
+    ]
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections, labels=labels)
+    ```
+
+![custom-label-annotation](https://media.roboflow.com/supervision_detect_and_annotate_example_2.png)
+
+## Annotate Image with Segmentations
+
+If you are running the segmentation model
+[`sv.MaskAnnotator`](/latest/annotators/#supervision.annotators.core.MaskAnnotator)
+is a drop-in replacement for
+[`sv.BoundingBoxAnnotator`](/latest/annotators/#supervision.annotators.core.BoundingBoxAnnotator)
+that will allow you to draw masks instead of boxes.
+
+=== "Inference"
+
+    ```python
+    import cv2
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8n-seg-640")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model.infer(image)[0]
+    detections = sv.Detections.from_inference(results)
+
+    mask_annotator = sv.MaskAnnotator()
+    label_annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER_OF_MASS)
+
+    annotated_image = mask_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+=== "Ultralytics"
+
+    ```python
+    import cv2
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8n-seg.pt")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model(image)[0]
+    detections = sv.Detections.from_ultralytics(results)
+
+    mask_annotator = sv.MaskAnnotator()
+    label_annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER_OF_MASS)
+
+    annotated_image = mask_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+=== "Transformers"
+
+    ```python
+    import torch
+    import supervision as sv
+    from PIL import Image
+    from transformers import DetrImageProcessor, DetrForSegmentation
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50-panoptic")
+    model = DetrForSegmentation.from_pretrained("facebook/detr-resnet-50-panoptic")
+
+    image = Image.open(<SOURCE_IMAGE_PATH>)
+    inputs = processor(images=image, return_tensors="pt")
+
+    with torch.no_grad():
+        outputs = model(**inputs)
+
+    width, height = image.size
+    target_size = torch.tensor([[height, width]])
+    results = processor.post_process_segmentation(
+        outputs=outputs, target_sizes=target_size)[0]
+    detections = sv.Detections.from_transformers(
+        transformers_results=results,
+        id2label=model.config.id2label)
+
+    mask_annotator = sv.MaskAnnotator()
+    label_annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER_OF_MASS)
+
+    labels = [
+        f"{class_name} {confidence:.2f}"
+        for class_name, confidence
+        in zip(detections['class_name'], detections.confidence)
+    ]
+
+    annotated_image = mask_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections, labels=labels)
+    ```
+
+![segmentation-annotation](https://media.roboflow.com/supervision_detect_and_annotate_example_3.png)
@@ -0,0 +1,326 @@
+---
+comments: true
+status: new
+---
+
+# Detect Small Objects
+
+This guide shows how to detect small objects
+with the [Inference](https://github.com/roboflow/inference),
+[Ultralytics](https://github.com/ultralytics/ultralytics) or
+[Transformers](https://github.com/huggingface/transformers) packages using
+[`InferenceSlicer`](/latest/detection/tools/inference_slicer/#supervision.detection.tools.inference_slicer.InferenceSlicer).
+
+<video controls>
+    <source src="https://media.roboflow.com/supervision_detect_small_objects_example.mp4" type="video/mp4">
+</video>
+
+## Baseline Detection
+
+Small object detection in high-resolution images presents challenges due to the objects'
+size relative to the image resolution.
+
+=== "Inference"
+
+    ```python
+    import cv2
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8x-640")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model.infer(image)[0]
+    detections = sv.Detections.from_inference(results)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+=== "Ultralytics"
+
+    ```python
+    import cv2
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8x.pt")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model(image)[0]
+    detections = sv.Detections.from_ultralytics(results)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+=== "Transformers"
+
+    ```python
+    import torch
+    import supervision as sv
+    from PIL import Image
+    from transformers import DetrImageProcessor, DetrForSegmentation
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForSegmentation.from_pretrained("facebook/detr-resnet-50")
+
+    image = Image.open(<SOURCE_IMAGE_PATH>)
+    inputs = processor(images=image, return_tensors="pt")
+
+    with torch.no_grad():
+        outputs = model(**inputs)
+
+    width, height = image_slice.size
+    target_size = torch.tensor([[width, height]])
+    results = processor.post_process_object_detection(
+        outputs=outputs, target_sizes=target_size)[0]
+    detections = sv.Detections.from_transformers(results)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    labels = [
+        model.config.id2label[class_id]
+        for class_id
+        in detections.class_id
+    ]
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections, labels=labels)
+    ```
+
+![basic-detection](https://media.roboflow.com/supervision_detect_small_objects_example_1.png)
+
+## Input Resolution
+
+Modifying the input resolution of images before detection can enhance small object
+identification at the cost of processing speed and increased memory usage. This method
+is less effective for ultra-high-resolution images (4K and above).
+
+=== "Inference"
+
+    ```{ .py hl_lines="5" }
+    import cv2
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8x-1280")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model.infer(image)[0]
+    detections = sv.Detections.from_inference(results)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+=== "Ultralytics"
+
+    ```{ .py hl_lines="7" }
+    import cv2
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8x.pt")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+    results = model(image, imgsz=1280)[0]
+    detections = sv.Detections.from_ultralytics(results)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+![detection-with-high-input-resolution](https://media.roboflow.com/supervision_detect_small_objects_example_2.png)
+
+## Inference Slicer
+
+[`InferenceSlicer`](/latest/detection/tools/inference_slicer/#supervision.detection.tools.inference_slicer.InferenceSlicer)
+processes high-resolution images by dividing them into smaller segments, detecting
+objects within each, and aggregating the results.
+
+<video controls>
+    <source src="https://media.roboflow.com/supervision_detect_small_objects_example_2.mp4" type="video/mp4">
+</video>
+
+=== "Inference"
+
+    ```{ .py hl_lines="9-14" }
+    import cv2
+    import numpy as np
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8x-640")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+
+    def callback(image_slice: np.ndarray) -> sv.Detections:
+        results = model.infer(image_slice)[0]
+        return sv.Detections.from_inference(results)
+
+    slicer = sv.InferenceSlicer(callback = callback)
+    detections = slicer(image)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+=== "Ultralytics"
+
+    ```{ .py hl_lines="9-14" }
+    import cv2
+    import numpy as np
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8x.pt")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+
+    def callback(image_slice: np.ndarray) -> sv.Detections:
+        result = model(image_slice)[0]
+        return sv.Detections.from_ultralytics(result)
+
+    slicer = sv.InferenceSlicer(callback = callback)
+    detections = slicer(image)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+=== "Transformers"
+
+    ```{ .py hl_lines="13-28" }
+    import cv2
+    import torch
+    import numpy as np
+    import supervision as sv
+    from PIL import Image
+    from transformers import DetrImageProcessor, DetrForObjectDetection
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+
+    def callback(image_slice: np.ndarray) -> sv.Detections:
+        image_slice = cv2.cvtColor(image_slice, cv2.COLOR_BGR2RGB)
+        image_slice = Image.fromarray(image_slice)
+        inputs = processor(images=image_slice, return_tensors="pt")
+
+        with torch.no_grad():
+            outputs = model(**inputs)
+
+        width, height = image_slice.size
+        target_size = torch.tensor([[width, height]])
+        results = processor.post_process_object_detection(
+            outputs=outputs, target_sizes=target_size)[0]
+        return sv.Detections.from_transformers(results)
+
+    slicer = sv.InferenceSlicer(callback = callback)
+    detections = slicer(image)
+
+    bounding_box_annotator = sv.BoundingBoxAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    labels = [
+        model.config.id2label[class_id]
+        for class_id
+        in detections.class_id
+    ]
+
+    annotated_image = bounding_box_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections, labels=labels)
+    ```
+
+![detection-with-inference-slicer](https://media.roboflow.com/supervision_detect_small_objects_example_3.png)
+
+## Small Object Segmentation
+
+[`InferenceSlicer`](/latest/detection/tools/inference_slicer/#supervision.detection.tools.inference_slicer.InferenceSlicer) can perform segmentation tasks too.
+
+=== "Inference"
+
+    ```{ .py hl_lines="6 16 19-20" }
+    import cv2
+    import numpy as np
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8x-seg-640")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+
+    def callback(image_slice: np.ndarray) -> sv.Detections:
+        results = model.infer(image_slice)[0]
+        return sv.Detections.from_inference(results)
+
+    slicer = sv.InferenceSlicer(callback = callback)
+    detections = slicer(image)
+
+    mask_annotator = sv.MaskAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = mask_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+=== "Ultralytics"
+
+    ```{ .py hl_lines="6 16 19-20" }
+    import cv2
+    import numpy as np
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8x-seg.pt")
+    image = cv2.imread(<SOURCE_IMAGE_PATH>)
+
+    def callback(image_slice: np.ndarray) -> sv.Detections:
+        result = model(image_slice)[0]
+        return sv.Detections.from_ultralytics(result)
+
+    slicer = sv.InferenceSlicer(callback = callback)
+    detections = slicer(image)
+
+    mask_annotator = sv.MaskAnnotator()
+    label_annotator = sv.LabelAnnotator()
+
+    annotated_image = mask_annotator.annotate(
+        scene=image, detections=detections)
+    annotated_image = label_annotator.annotate(
+        scene=annotated_image, detections=detections)
+    ```
+
+![detection-with-inference-slicer](https://media.roboflow.com/supervision-docs/inference-slicer-segmentation-example.png)
@@ -0,0 +1,298 @@
+---
+comments: true
+---
+
+# Save Detections
+
+Supervision enables an easy way to save detections in .CSV and .JSON files for offline
+processing. This guide demonstrates how to perform video inference using the
+[Inference](https://github.com/roboflow/inference),
+[Ultralytics](https://github.com/ultralytics/ultralytics) or
+[Transformers](https://github.com/huggingface/transformers) packages and save their results with
+[`sv.CSVSink`](/latest/detection/tools/save_detections/#supervision.detection.tools.csv_sink.CSVSink) and
+[`sv.JSONSink`](/latest/detection/tools/save_detections/#supervision.detection.tools.csv_sink.JSONSink).
+
+## Run Detection
+
+First, you'll need to obtain predictions from your object detection or segmentation
+model. You can learn more on this topic in our
+[How to Detect and Annotate](/latest/how_to/detect_and_annotate.md) guide.
+
+=== "Inference"
+
+    ```python
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8n-640")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    for frame in frames_generator:
+
+        results = model.infer(image)[0]
+        detections = sv.Detections.from_inference(results)
+    ```
+
+=== "Ultralytics"
+
+    ```python
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8n.pt")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    for frame in frames_generator:
+
+        results = model(frame)[0]
+        detections = sv.Detections.from_ultralytics(results)
+    ```
+
+=== "Transformers"
+
+    ```python
+    import torch
+    import supervision as sv
+    from transformers import DetrImageProcessor, DetrForObjectDetection
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    for frame in frames_generator:
+
+        frame = sv.cv2_to_pillow(frame)
+        inputs = processor(images=frame, return_tensors="pt")
+
+        with torch.no_grad():
+            outputs = model(**inputs)
+
+        width, height = frame.size
+        target_size = torch.tensor([[height, width]])
+        results = processor.post_process_object_detection(
+            outputs=outputs, target_sizes=target_size)[0]
+        detections = sv.Detections.from_transformers(results)
+    ```
+
+## Save Detections as CSV
+
+To save detections to a `.CSV` file, open our
+[`sv.CSVSink`](/latest/detection/tools/save_detections/#supervision.detection.tools.csv_sink.CSVSink)
+and then pass the
+[`sv.Detections`](/latest/detection/core/#supervision.detection.core.Detections)
+object resulting from the inference to it. Its fields are parsed and saved on disk.
+
+=== "Inference"
+
+    ```{ .py hl_lines="7 12" }
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8n-640")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    with sv.CSVSink(<TARGET_CSV_PATH>) as sink:
+        for frame in frames_generator:
+
+            results = model.infer(image)[0]
+            detections = sv.Detections.from_inference(results)
+            sink.append(detections, {})
+    ```
+
+=== "Ultralytics"
+
+    ```{ .py hl_lines="7 12" }
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8n.pt")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    with sv.CSVSink(<TARGET_CSV_PATH>) as sink:
+        for frame in frames_generator:
+
+            results = model(frame)[0]
+            detections = sv.Detections.from_ultralytics(results)
+            sink.append(detections, {})
+    ```
+
+=== "Transformers"
+
+    ```{ .py hl_lines="9 23" }
+    import torch
+    import supervision as sv
+    from transformers import DetrImageProcessor, DetrForObjectDetection
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    with sv.CSVSink(<TARGET_CSV_PATH>) as sink:
+        for frame in frames_generator:
+
+            frame = sv.cv2_to_pillow(frame)
+            inputs = processor(images=frame, return_tensors="pt")
+
+            with torch.no_grad():
+                outputs = model(**inputs)
+
+            width, height = frame.size
+            target_size = torch.tensor([[height, width]])
+            results = processor.post_process_object_detection(
+                outputs=outputs, target_sizes=target_size)[0]
+            detections = sv.Detections.from_transformers(results)
+            sink.append(detections, {})
+    ```
+
+| x_min   | y_min    | x_max   | y_max    | class_id | confidence | tracker_id | class_name |
+|---------|----------|---------|----------|----------|------------|------------|------------|
+| 2941.14 | 1269.31  | 3220.77 | 1500.67  | 2        | 0.8517     |            | car        |
+| 944.889 | 899.641  | 1235.42 | 1308.80  | 7        | 0.6752     |            | truck      |
+| 1439.78 | 1077.79  | 1621.27 | 1231.40  | 2        | 0.6450     |            | car        |
+
+## Custom Fields
+
+Besides regular fields in
+[`sv.Detections`](/latest/detection/core/#supervision.detection.core.Detections),
+[`sv.CSVSink`](/latest/detection/tools/save_detections/#supervision.detection.tools.csv_sink.CSVSink)
+also allows you to add custom information to each row, which can be passed via the
+`custom_data` dictionary. Let's utilize this feature to save information about the
+frame index from which the detections originate.
+
+=== "Inference"
+
+    ```{ .py hl_lines="8 12" }
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8n-640")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    with sv.CSVSink(<TARGET_CSV_PATH>) as sink:
+        for frame_index, frame in enumerate(frames_generator):
+
+            results = model.infer(image)[0]
+            detections = sv.Detections.from_inference(results)
+            sink.append(detections, {"frame_index": frame_index})
+    ```
+
+=== "Ultralytics"
+
+    ```{ .py hl_lines="8 12" }
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8n.pt")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    with sv.CSVSink(<TARGET_CSV_PATH>) as sink:
+        for frame_index, frame in enumerate(frames_generator):
+
+            results = model(frame)[0]
+            detections = sv.Detections.from_ultralytics(results)
+            sink.append(detections, {"frame_index": frame_index})
+    ```
+
+=== "Transformers"
+
+    ```{ .py hl_lines="10 23" }
+    import torch
+    import supervision as sv
+    from transformers import DetrImageProcessor, DetrForObjectDetection
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    with sv.CSVSink(<TARGET_CSV_PATH>) as sink:
+        for frame_index, frame in enumerate(frames_generator):
+
+            frame = sv.cv2_to_pillow(frame)
+            inputs = processor(images=frame, return_tensors="pt")
+
+            with torch.no_grad():
+                outputs = model(**inputs)
+
+            width, height = frame.size
+            target_size = torch.tensor([[height, width]])
+            results = processor.post_process_object_detection(
+                outputs=outputs, target_sizes=target_size)[0]
+            detections = sv.Detections.from_transformers(results)
+            sink.append(detections, {"frame_index": frame_index})
+    ```
+
+| x_min   | y_min    | x_max   | y_max    | class_id | confidence | tracker_id | class_name | frame_index |
+|---------|----------|---------|----------|----------|------------|------------|------------|-------------|
+| 2941.14 | 1269.31  | 3220.77 | 1500.67  | 2        | 0.8517     |            | car        | 0           |
+| 944.889 | 899.641  | 1235.42 | 1308.80  | 7        | 0.6752     |            | truck      | 0           |
+| 1439.78 | 1077.79  | 1621.27 | 1231.40  | 2        | 0.6450     |            | car        | 0           |
+
+## Save Detections as JSON
+
+If you prefer to save the result in a `.JSON` file instead of a `.CSV` file, all you
+need to do is replace
+[`sv.CSVSink`](/latest/detection/tools/save_detections/#supervision.detection.tools.csv_sink.CSVSink)
+with
+[`sv.JSONSink`](/latest/detection/tools/save_detections/#supervision.detection.tools.csv_sink.JSONSink).
+
+=== "Inference"
+
+    ```{ .py hl_lines="7" }
+    import supervision as sv
+    from inference import get_model
+
+    model = get_model(model_id="yolov8n-640")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    with sv.JSONSink(<TARGET_CSV_PATH>) as sink:
+        for frame_index, frame in enumerate(frames_generator):
+
+            results = model.infer(image)[0]
+            detections = sv.Detections.from_inference(results)
+            sink.append(detections, {"frame_index": frame_index})
+    ```
+
+=== "Ultralytics"
+
+    ```{ .py hl_lines="7" }
+    import supervision as sv
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8n.pt")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    with sv.JSONSink(<TARGET_CSV_PATH>) as sink:
+        for frame_index, frame in enumerate(frames_generator):
+
+            results = model(frame)[0]
+            detections = sv.Detections.from_ultralytics(results)
+            sink.append(detections, {"frame_index": frame_index})
+    ```
+
+=== "Transformers"
+
+    ```{ .py hl_lines="9" }
+    import torch
+    import supervision as sv
+    from transformers import DetrImageProcessor, DetrForObjectDetection
+
+    processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+    model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+    frames_generator = sv.get_video_frames_generator(<SOURCE_VIDEO_PATH>)
+
+    with sv.JSONSink(<TARGET_CSV_PATH>) as sink:
+        for frame_index, frame in enumerate(frames_generator):
+
+            frame = sv.cv2_to_pillow(frame)
+            inputs = processor(images=frame, return_tensors="pt")
+
+            with torch.no_grad():
+                outputs = model(**inputs)
+
+            width, height = frame.size
+            target_size = torch.tensor([[height, width]])
+            results = processor.post_process_object_detection(
+                outputs=outputs, target_sizes=target_size)[0]
+            detections = sv.Detections.from_transformers(results)
+            sink.append(detections, {"frame_index": frame_index})
+    ```
@@ -30,7 +30,7 @@ We write your reusable computer vision tools. Whether you need to load your data

 ## 💻 Install

-You can install `supervision` with pip in a
+You can install `supervision` in a
 [**Python>=3.8**](https://www.python.org/) environment.

 !!! example "pip install (recommended)"
@@ -49,6 +49,24 @@ You can install `supervision` with pip in a
        pip install "supervision[desktop]"
        ```

+!!! example "conda/mamba install"
+
+    === "conda"
+
+        [![conda-recipe](https://img.shields.io/badge/recipe-supervision-green.svg)](https://anaconda.org/conda-forge/supervision) [![conda-downloads](https://img.shields.io/conda/dn/conda-forge/supervision.svg)](https://anaconda.org/conda-forge/supervision) [![conda-version](https://img.shields.io/conda/vn/conda-forge/supervision.svg)](https://anaconda.org/conda-forge/supervision) [![conda-platforms](https://img.shields.io/conda/pn/conda-forge/supervision.svg)](https://anaconda.org/conda-forge/supervision)
+
+        ```bash
+        conda install -c conda-forge supervision
+        ```
+
+    === "mamba"
+
+        [![mamba-recipe](https://img.shields.io/badge/recipe-supervision-green.svg)](https://anaconda.org/conda-forge/supervision) [![mamba-downloads](https://img.shields.io/conda/dn/conda-forge/supervision.svg)](https://anaconda.org/conda-forge/supervision) [![mamba-version](https://img.shields.io/conda/vn/conda-forge/supervision.svg)](https://anaconda.org/conda-forge/supervision) [![mamba-platforms](https://img.shields.io/conda/pn/conda-forge/supervision.svg)](https://anaconda.org/conda-forge/supervision)
+
+        ```bash
+        mamba install -c conda-forge supervision
+        ```
+
 !!! example "git clone (for development)"

    === "virtualenv"
@@ -88,7 +106,6 @@ You can install `supervision` with pip in a
        poetry install --extras "desktop"
        ```

-
 ## 🚀 Quickstart

 <div class="grid cards" markdown>
@@ -109,6 +126,14 @@ You can install `supervision` with pip in a

    [:octicons-arrow-right-24: Tutorial](how_to/track_objects.md)

+-   __Detect Small Objects__
+
+    ---
+
+    Learn how to detect small objects in images
+
+    [:octicons-arrow-right-24: Tutorial](how_to/detect_small_objects.md)
+
 -   > __Count Objects Crossing Line__

    ---
@@ -0,0 +1,97 @@
+---
+comments: true
+status: new
+---
+
+# Annotators
+
+=== "VertexAnnotator"
+
+    ```python
+    import supervision as sv
+
+    image = ...
+    key_points = sv.KeyPoints(...)
+
+    vertex_annotator = sv.VertexAnnotator(
+        color=sv.Color.GREEN,
+        radius=10
+    )
+    annotated_frame = vertex_annotator.annotate(
+        scene=image.copy(),
+        key_points=key_points
+    )
+    ```
+
+    <div class="result" markdown>
+
+    ![vertex-annotator-example](https://media.roboflow.com/supervision-annotator-examples/vertex-annotator-example.png){ align=center width="800" }
+
+    </div>
+
+=== "EdgeAnnotator"
+
+    ```python
+    import supervision as sv
+
+    image = ...
+    key_points = sv.KeyPoints(...)
+
+    edge_annotator = sv.EdgeAnnotator(
+        color=sv.Color.GREEN,
+        thickness=5
+    )
+    annotated_frame = edge_annotator.annotate(
+        scene=image.copy(),
+        key_points=key_points
+    )
+    ```
+
+    <div class="result" markdown>
+
+    ![edge-annotator-example](https://media.roboflow.com/supervision-annotator-examples/edge-annotator-example.png){ align=center width="800" }
+
+    </div>
+
+=== "VertexLabelAnnotator"
+
+    ```python
+    import supervision as sv
+
+    image = ...
+    key_points = sv.KeyPoints(...)
+
+    vertex_label_annotator = sv.VertexLabelAnnotator(
+        color=sv.Color.GREEN,
+        text_color=sv.Color.BLACK,
+        border_radius=5
+    )
+    annotated_frame = vertex_label_annotator.annotate(
+        scene=image.copy(),
+        key_points=key_points
+    )
+    ```
+
+    <div class="result" markdown>
+
+    ![vertex-label-annotator-example](https://media.roboflow.com/supervision-annotator-examples/vertex-label-annotator-example.png){ align=center width="800" }
+
+    </div>
+
+<div class="md-typeset">
+  <h2><a href="#supervision.keypoint.annotators.VertexAnnotator">VertexAnnotator</a></h2>
+</div>
+
+:::supervision.keypoint.annotators.VertexAnnotator
+
+<div class="md-typeset">
+  <h2><a href="#supervision.keypoint.annotators.EdgeAnnotator">EdgeAnnotator</a></h2>
+</div>
+
+:::supervision.keypoint.annotators.EdgeAnnotator
+
+<div class="md-typeset">
+  <h2><a href="#supervision.keypoint.annotators.VertexLabelAnnotator">VertexLabelAnnotator</a></h2>
+</div>
+
+:::supervision.keypoint.annotators.VertexLabelAnnotator
@@ -0,0 +1,8 @@
+---
+comments: true
+status: new
+---
+
+# Keypoint Detection
+
+:::supervision.keypoint.core.KeyPoints
@@ -1,22 +0,0 @@
---
-comments: true
---
-
-#  Detection Metrics
-
-!!! warning
-
-    Evaluation API is still fluid and may change. If you use Evaluation API in your project until further notice, freeze the
-    `supervision` version in your `requirements.txt` or `setup.py`.
-
-<div class="md-typeset">
-    <h2><a href="#supervision.metrics.detection.ConfusionMatrix">ConfusionMatrix</a></h2>
-</div>
-
-:::supervision.metrics.detection.ConfusionMatrix
-
-<div class="md-typeset">
-    <h2><a href="#supervision.annotators.core.MeanAveragePrecision">MeanAveragePrecision</a></h2>
-</div>
-
-:::supervision.metrics.detection.MeanAveragePrecision
@@ -19,6 +19,9 @@
      <p class="card repo-card" data-url="/develop/notebooks/object-tracking" data-name="Object Tracking"  data-labels="TRACKING, ANNOTATOR" data-version="v0.18.0" data-author="nickherrig"></p>
      <p class="card repo-card" data-url="/develop/notebooks/occupancy_analytics" data-name="Analyzing Zone Occupancy"  data-labels="ANNOTATOR,DETECTION,ZONES" data-version="v0.19.0" data-author="stellasphere"></p>
      <p class="card repo-card" data-url="/develop/notebooks/evaluating-alignment-of-text-to-image-diffusion-models" data-name="Evaluating Alignment of Text-to-image Diffusion Models"  data-labels="ANNOTATORS,YOLO WORLD" data-version="v0.19.0rc5" data-author="iamhatesz"></p>
+      <p class="card repo-card" data-url="/develop/notebooks/serialise-detections-to-csv" data-name="Serialise Detections to a CSV File"  data-labels="DETECTIONS,CSV SINK,INFERENCE" data-version="v0.21.0" data-author="onuralpszr"></p>
+      <p class="card repo-card" data-url="/develop/notebooks/serialise-detections-to-json" data-name="Serialise Detections to a JSON File"  data-labels="DETECTIONS,JSON SINK,INFERENCE" data-version="v0.21.0" data-author="onuralpszr"></p>
+
    </div>
  </div>
 </section>
@@ -1,6 +1,5 @@
 ---
 comments: true
-status: new
 ---

 # ByteTrack
@@ -0,0 +1,65 @@
+---
+comments: true
+---
+
+# Draw Utils
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.utils.draw_line">draw_line</a></h2>
+</div>
+
+:::supervision.draw.utils.draw_line
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.utils.draw_rectangle">draw_rectangle</a></h2>
+</div>
+
+:::supervision.draw.utils.draw_rectangle
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.utils.draw_filled_rectangle">draw_filled_rectangle</a></h2>
+</div>
+
+:::supervision.draw.utils.draw_filled_rectangle
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.utils.draw_polygon">draw_polygon</a></h2>
+</div>
+
+:::supervision.draw.utils.draw_polygon
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.utils.draw_text">draw_text</a></h2>
+</div>
+
+:::supervision.draw.utils.draw_text
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.utils.draw_image">draw_image</a></h2>
+</div>
+
+:::supervision.draw.utils.draw_image
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.utils.calculate_optimal_text_scale">calculate_optimal_text_scale</a></h2>
+</div>
+
+:::supervision.draw.utils.calculate_optimal_text_scale
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.utils.calculate_optimal_line_thickness">calculate_optimal_line_thickness</a></h2>
+</div>
+
+:::supervision.draw.utils.calculate_optimal_line_thickness
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.color.Color">Color</a></h2>
+</div>
+
+:::supervision.draw.color.Color
+
+<div class="md-typeset">
+    <h2><a href="#supervision.draw.color.ColorPalette">ColorPalette</a></h2>
+</div>
+
+:::supervision.draw.color.ColorPalette
@@ -5,7 +5,7 @@ comments: true
 # File Utils

 <div class="md-typeset">
-  <h2>list_files_with_extensions</h2>
+    <h2><a href="#supervision.utils.file.list_files_with_extensions">list_files_with_extensions</a></h2>
 </div>

 :::supervision.utils.file.list_files_with_extensions
@@ -0,0 +1,15 @@
+---
+comments: true
+---
+
+<div class="md-typeset">
+    <h2><a href="#supervision.geometry.core.utils.get_polygon_center">get_polygon_center</a></h2>
+</div>
+
+:::supervision.geometry.utils.get_polygon_center
+
+<div class="md-typeset">
+    <h2><a href="#supervision.geometry.core.Position">Position</a></h2>
+</div>
+
+:::supervision.geometry.core.Position
@@ -1,30 +1,41 @@
 ---
 comments: true
-status: new
 ---

 # Image Utils

 <div class="md-typeset">
-  <h2>ImageSink</h2>
-</div>
-
-:::supervision.utils.image.ImageSink
-
-<div class="md-typeset">
-  <h2>crop_image</h2>
+    <h2><a href="#supervision.utils.image.crop_image">crop_image</a></h2>
 </div>

 :::supervision.utils.image.crop_image

 <div class="md-typeset">
-  <h2>resize_image</h2>
+    <h2><a href="#supervision.utils.image.scale_image">scale_image</a></h2>
+</div>
+
+:::supervision.utils.image.scale_image
+
+<div class="md-typeset">
+    <h2><a href="#supervision.utils.image.resize_image">resize_image</a></h2>
 </div>

 :::supervision.utils.image.resize_image

 <div class="md-typeset">
-  <h2>place_image</h2>
+    <h2><a href="#supervision.utils.image.letterbox_image">letterbox_image</a></h2>
 </div>

-:::supervision.utils.image.place_image
+:::supervision.utils.image.letterbox_image
+
+<div class="md-typeset">
+    <h2><a href="#supervision.utils.image.overlay_image">overlay_image</a></h2>
+</div>
+
+:::supervision.utils.image.overlay_image
+
+<div class="md-typeset">
+    <h2><a href="#supervision.utils.image.ImageSink">ImageSink</a></h2>
+</div>
+
+:::supervision.utils.image.ImageSink
@@ -0,0 +1,17 @@
+---
+comments: true
+---
+
+# Iterables Utils
+
+<div class="md-typeset">
+    <h2><a href="#supervision.utils.iterables.create_batches">create_batches</a></h2>
+</div>
+
+:::supervision.utils.iterables.create_batches
+
+<div class="md-typeset">
+    <h2><a href="#supervision.utils.iterables.fill">fill</a></h2>
+</div>
+
+:::supervision.utils.iterables.fill
@@ -5,14 +5,13 @@ comments: true
 # Notebooks Utils

 <div class="md-typeset">
-  <h2>plot_image</h2>
+    <h2><a href="#supervision.utils.notebook.plot_image">plot_image</a></h2>
 </div>

 :::supervision.utils.notebook.plot_image

 <div class="md-typeset">
-  <h2>## plot_images_grid
-</h2>
+    <h2><a href="#supervision.utils.notebook.plot_images_grid">plot_images_grid</a></h2>
 </div>

 :::supervision.utils.notebook.plot_images_grid
@@ -5,31 +5,31 @@ comments: true
 # Video Utils

 <div class="md-typeset">
-  <h2>VideoInfo</h2>
+    <h2><a href="#supervision.utils.video.VideoInfo">VideoInfo</a></h2>
 </div>

 :::supervision.utils.video.VideoInfo

 <div class="md-typeset">
-  <h2>VideoSink</h2>
+    <h2><a href="#supervision.utils.video.VideoSink">VideoSink</a></h2>
 </div>

 :::supervision.utils.video.VideoSink

 <div class="md-typeset">
-  <h2>FPSMonitor</h2>
+    <h2><a href="#supervision.utils.video.FPSMonitor">FPSMonitor</a></h2>
 </div>

 :::supervision.utils.video.FPSMonitor

 <div class="md-typeset">
-  <h2>get_video_frames_generator</h2>
+    <h2><a href="#supervision.utils.video.get_video_frames_generator">get_video_frames_generator</a></h2>
 </div>

 :::supervision.utils.video.get_video_frames_generator

 <div class="md-typeset">
-  <h2>process_video</h2>
+    <h2><a href="#supervision.utils.video.process_video">process_video</a></h2>
 </div>

 :::supervision.utils.video.process_video
@@ -1,31 +1,12 @@
 # Examples

-This repository is packed with real-world use-cases, provided through Python scripts or
-interactive notebooks. Browse through to understand how the Supervision library
-interfaces with diverse applications.
+Here, you'll find end-to-end examples that show how to solve common computer vision problems using Supervision.
+
+For more information and examples, visit our [documentation](https://supervision.roboflow.com/develop/annotators/) and explore our [how-to guides](https://supervision.roboflow.com/develop/how_to/detect_and_annotate/) and [cookbooks](https://supervision.roboflow.com/develop/cookbooks/). Join our [Discord](https://discord.com/invite/GbfgXGJ8Bk) and meet other Supervision power users!

 - [tracking](./tracking) by [@SkalskiP](https://github.com/SkalskiP)
 - [count people in zone](./count_people_in_zone) by [@SkalskiP](https://github.com/SkalskiP)
 - [traffic analysis](./traffic_analysis) by [@SkalskiP](https://github.com/SkalskiP)
 - [speed estimation](./speed_estimation) by [@SkalskiP](https://github.com/SkalskiP)
+- [time in zone](./time_in_zone) by [@SkalskiP](https://github.com/SkalskiP)
 - [heatmap and track](./heatmap_and_track/) by [@HinePo](https://github.com/HinePo)
-
-## Contributing
-
-We welcome contributions from the community in the form of examples, applications, and
-guides. To contribute, please follow these steps:
-
-1. Create a pull request (PR) with the `[Example]` prefix in the title, adding your
-project folder to the `examples/` directory in the repository.
-2. Confirm your project aligns with the following standards:
-   - Incorporates the `supervision` package.
-   - Provides a `README.md` file, detailing the instructions to execute the project.
-   - Showcases visual results, demonstrating the app's functionality.
-   - Avoids adding large assets or dependencies unless absolutely necessary.
-   - The contributor is expected to provide support for issues related to their
-examples.
-   - In case the presented model has licensing complications, kindly specify them to
-circumvent potential misunderstandings.
-
-For inquiries or concerns about these prerequisites, feel free to raise a PR. We are
-committed to assist and guide you.
@@ -1,5 +1,8 @@
 # count people in zone

+[![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/how-to-detect-and-count-objects-in-polygon-zone.ipynb)
+[![YouTube](https://badges.aleen42.com/src/youtube.svg)](https://www.youtube.com/watch?v=l_kf9CfZ_8M)
+
 ## 👋 hello

 This demo is a video analysis tool that counts and highlights objects in specific zones
@@ -38,15 +38,15 @@ def initiate_annotators(
 ) -> Tuple[
    List[sv.PolygonZone], List[sv.PolygonZoneAnnotator], List[sv.BoundingBoxAnnotator]
 ]:
-    line_thickness = sv.calculate_dynamic_line_thickness(resolution_wh=resolution_wh)
-    text_scale = sv.calculate_dynamic_text_scale(resolution_wh=resolution_wh)
+    line_thickness = sv.calculate_optimal_line_thickness(resolution_wh=resolution_wh)
+    text_scale = sv.calculate_optimal_text_scale(resolution_wh=resolution_wh)

    zones = []
    zone_annotators = []
    box_annotators = []

    for index, polygon in enumerate(polygons):
-        zone = sv.PolygonZone(polygon=polygon, frame_resolution_wh=resolution_wh)
+        zone = sv.PolygonZone(polygon=polygon)
        zone_annotator = sv.PolygonZoneAnnotator(
            zone=zone,
            color=COLORS.by_idx(index),
@@ -1,5 +1,5 @@
 gdown
-inference
-supervision
+inference==0.9.17
+supervision>=0.20.0
 tqdm
 ultralytics
@@ -36,15 +36,15 @@ def initiate_annotators(
 ) -> Tuple[
    List[sv.PolygonZone], List[sv.PolygonZoneAnnotator], List[sv.BoundingBoxAnnotator]
 ]:
-    line_thickness = sv.calculate_dynamic_line_thickness(resolution_wh=resolution_wh)
-    text_scale = sv.calculate_dynamic_text_scale(resolution_wh=resolution_wh)
+    line_thickness = sv.calculate_optimal_line_thickness(resolution_wh=resolution_wh)
+    text_scale = sv.calculate_optimal_text_scale(resolution_wh=resolution_wh)

    zones = []
    zone_annotators = []
    box_annotators = []

    for index, polygon in enumerate(polygons):
-        zone = sv.PolygonZone(polygon=polygon, frame_resolution_wh=resolution_wh)
+        zone = sv.PolygonZone(polygon=polygon)
        zone_annotator = sv.PolygonZoneAnnotator(
            zone=zone,
            color=COLORS.by_idx(index),
@@ -1,2 +1,2 @@
-supervision[assets]
+supervision[assets]==0.19.0
 ultralytics
@@ -98,10 +98,10 @@ if __name__ == "__main__":
        frame_rate=video_info.fps, track_thresh=args.confidence_threshold
    )

-    thickness = sv.calculate_dynamic_line_thickness(
+    thickness = sv.calculate_optimal_line_thickness(
        resolution_wh=video_info.resolution_wh
    )
-    text_scale = sv.calculate_dynamic_text_scale(resolution_wh=video_info.resolution_wh)
+    text_scale = sv.calculate_optimal_text_scale(resolution_wh=video_info.resolution_wh)
    bounding_box_annotator = sv.BoundingBoxAnnotator(thickness=thickness)
    label_annotator = sv.LabelAnnotator(
        text_scale=text_scale,
@@ -116,9 +116,7 @@ if __name__ == "__main__":

    frame_generator = sv.get_video_frames_generator(source_path=args.source_video_path)

-    polygon_zone = sv.PolygonZone(
-        polygon=SOURCE, frame_resolution_wh=video_info.resolution_wh
-    )
+    polygon_zone = sv.PolygonZone(polygon=SOURCE)
    view_transformer = ViewTransformer(source=SOURCE, target=TARGET)

    coordinates = defaultdict(lambda: deque(maxlen=video_info.fps))
@@ -1,6 +1,6 @@
-supervision==0.18.0rc1
-tqdm==4.66.1
+supervision>=0.20.0
+tqdm==4.66.3
 requests
 ultralytics==8.0.237
 super-gradients==3.5.0
-inference==0.9.8
+inference==0.9.17
@@ -76,10 +76,10 @@ if __name__ == "__main__":
        frame_rate=video_info.fps, track_thresh=args.confidence_threshold
    )

-    thickness = sv.calculate_dynamic_line_thickness(
+    thickness = sv.calculate_optimal_line_thickness(
        resolution_wh=video_info.resolution_wh
    )
-    text_scale = sv.calculate_dynamic_text_scale(resolution_wh=video_info.resolution_wh)
+    text_scale = sv.calculate_optimal_text_scale(resolution_wh=video_info.resolution_wh)
    bounding_box_annotator = sv.BoundingBoxAnnotator(thickness=thickness)
    label_annotator = sv.LabelAnnotator(
        text_scale=text_scale,
@@ -94,9 +94,7 @@ if __name__ == "__main__":

    frame_generator = sv.get_video_frames_generator(source_path=args.source_video_path)

-    polygon_zone = sv.PolygonZone(
-        polygon=SOURCE, frame_resolution_wh=video_info.resolution_wh
-    )
+    polygon_zone = sv.PolygonZone(polygon=SOURCE)
    view_transformer = ViewTransformer(source=SOURCE, target=TARGET)

    coordinates = defaultdict(lambda: deque(maxlen=video_info.fps))
@@ -77,10 +77,10 @@ if __name__ == "__main__":
        frame_rate=video_info.fps, track_thresh=args.confidence_threshold
    )

-    thickness = sv.calculate_dynamic_line_thickness(
+    thickness = sv.calculate_optimal_line_thickness(
        resolution_wh=video_info.resolution_wh
    )
-    text_scale = sv.calculate_dynamic_text_scale(resolution_wh=video_info.resolution_wh)
+    text_scale = sv.calculate_optimal_text_scale(resolution_wh=video_info.resolution_wh)
    bounding_box_annotator = sv.BoundingBoxAnnotator(thickness=thickness)
    label_annotator = sv.LabelAnnotator(
        text_scale=text_scale,
@@ -95,9 +95,7 @@ if __name__ == "__main__":

    frame_generator = sv.get_video_frames_generator(source_path=args.source_video_path)

-    polygon_zone = sv.PolygonZone(
-        polygon=SOURCE, frame_resolution_wh=video_info.resolution_wh
-    )
+    polygon_zone = sv.PolygonZone(polygon=SOURCE)
    view_transformer = ViewTransformer(source=SOURCE, target=TARGET)

    coordinates = defaultdict(lambda: deque(maxlen=video_info.fps))
@@ -0,0 +1,9 @@
+data/
+venv*/
+*.pt
+*.pth
+*.mp4
+*.mov
+*.png
+*.jpg
+*.jpeg
@@ -0,0 +1,264 @@
+# time in zone
+
+[![YouTube](https://badges.aleen42.com/src/youtube.svg)](https://www.youtube.com/watch?v=hAWpsIuem10)
+
+## 👋 hello
+
+Practical demonstration on leveraging computer vision for analyzing wait times and
+monitoring the duration that objects or individuals spend in predefined areas of video
+frames. This example project, perfect for retail analytics or traffic management
+applications.
+
+https://github.com/roboflow/supervision/assets/26109316/d051cc8a-dd15-41d4-aa36-d38b86334c39
+
+## 💻 install
+
+- clone repository and navigate to example directory
+
+  ```bash
+  git clone https://github.com/roboflow/supervision.git
+  cd supervision/examples/time_in_zone
+  ```
+
+- setup python environment and activate it [optional]
+
+  ```bash
+  python3 -m venv venv
+  source venv/bin/activate
+  ```
+
+- install required dependencies
+
+  ```bash
+  pip install -r requirements.txt
+  ```
+
+## 🛠 scripts
+
+### `download_from_youtube`
+
+This script allows you to download a video from YouTube.
+
+- `--url`: The full URL of the YouTube video you wish to download.
+- `--output_path` (optional): Specifies the directory where the video will be saved.
+- `--file_name` (optional): Sets the name of the saved video file.
+
+```bash
+python scripts/download_from_youtube.py \
+--url "https://www.youtube.com/watch?v=-8zyEwAa50Q" \
+--output_path "data/checkout" \
+--file_name "video.mp4"
+```
+
+```bash
+python scripts/download_from_youtube.py \
+--url "https://www.youtube.com/watch?v=MNn9qKG2UFI" \
+--output_path "data/traffic" \
+--file_name "video.mp4"
+```
+
+### `stream_from_file`
+
+This script allows you to stream video files from a directory. It's an awesome way to
+mock a live video stream for local testing. Video will be streamed in a loop under
+`rtsp://localhost:8554/live0.stream` URL. This script requires docker to be installed.
+
+- `--video_directory`: Directory containing video files to stream.
+- `--number_of_streams`: Number of video files to stream.
+
+```bash
+python scripts/stream_from_file.py \
+--video_directory "data/checkout" \
+--number_of_streams 1
+```
+
+```bash
+python scripts/stream_from_file.py \
+--video_directory "data/traffic" \
+--number_of_streams 1
+```
+
+### `draw_zones`
+
+If you want to test zone time in zone analysis on your own video, you can use this
+script to design custom zones and save results as a JSON file. The script will open a
+window where you can draw polygons on the source image or video file. The polygons will
+be saved as a JSON file.
+
+- `--source_path`: Path to the source image or video file for drawing polygons.
+- `--zone_configuration_path`: Path where the polygon annotations will be saved as a JSON file.
+
+
+- `enter` - finish drawing the current polygon.
+- `escape` - cancel drawing the current polygon.
+- `q` - quit the drawing window.
+- `s` - save zone configuration to a JSON file.
+
+```bash
+python scripts/draw_zones.py \
+--source_path "data/checkout/video.mp4" \
+--zone_configuration_path "data/checkout/config.json"
+```
+
+```bash
+python scripts/draw_zones.py \
+--source_path "data/traffic/video.mp4" \
+--zone_configuration_path "data/traffic/config.json"
+```
+
+https://github.com/roboflow/supervision/assets/26109316/9d514c9e-2a61-418b-ae49-6ac1ad6ae5ac
+
+## 🎬 video & stream processing
+
+### `inference_file_example`
+
+Script to run object detection on a video file using the Roboflow Inference model.
+
+  - `--zone_configuration_path`: Path to the zone configuration JSON file.
+  - `--source_video_path`: Path to the source video file.
+  - `--model_id`: Roboflow model ID.
+  - `--classes`: List of class IDs to track. If empty, all classes are tracked.
+  - `--confidence_threshold`: Confidence level for detections (`0` to `1`). Default is `0.3`.
+  - `--iou_threshold`: IOU threshold for non-max suppression. Default is `0.7`.
+
+```bash
+python inference_file_example.py \
+--zone_configuration_path "data/checkout/config.json" \
+--source_video_path "data/checkout/video.mp4" \
+--model_id "yolov8x-640" \
+--classes 0 \
+--confidence_threshold 0.3 \
+--iou_threshold 0.7
+```
+
+https://github.com/roboflow/supervision/assets/26109316/d051cc8a-dd15-41d4-aa36-d38b86334c39
+
+```bash
+python inference_file_example.py \
+--zone_configuration_path "data/traffic/config.json" \
+--source_video_path "data/traffic/video.mp4" \
+--model_id "yolov8x-640" \
+--classes 2 5 6 7 \
+--confidence_threshold 0.3 \
+--iou_threshold 0.7
+```
+
+https://github.com/roboflow/supervision/assets/26109316/5ec896d7-4b39-4426-8979-11e71666878b
+
+### `inference_stream_example`
+
+Script to run object detection on a video stream using the Roboflow Inference model.
+
+  - `--zone_configuration_path`: Path to the zone configuration JSON file.
+  - `--rtsp_url`: Complete RTSP URL for the video stream.
+  - `--model_id`: Roboflow model ID.
+  - `--classes`: List of class IDs to track. If empty, all classes are tracked.
+  - `--confidence_threshold`: Confidence level for detections (`0` to `1`). Default is `0.3`.
+  - `--iou_threshold`: IOU threshold for non-max suppression. Default is `0.7`.
+
+```bash
+python inference_stream_example.py \
+--zone_configuration_path "data/checkout/config.json" \
+--rtsp_url "rtsp://localhost:8554/live0.stream" \
+--model_id "yolov8x-640" \
+--classes 0 \
+--confidence_threshold 0.3 \
+--iou_threshold 0.7
+```
+
+```bash
+python inference_stream_example.py \
+--zone_configuration_path "data/traffic/config.json" \
+--rtsp_url "rtsp://localhost:8554/live0.stream" \
+--model_id "yolov8x-640" \
+--classes 2 5 6 7 \
+--confidence_threshold 0.3 \
+--iou_threshold 0.7
+```
+
+<details>
+<summary>👉 show ultralytics examples</summary>
+
+### `ultralytics_file_example`
+
+Script to run object detection on a video file using the Ultralytics YOLOv8 model.
+
+  - `--zone_configuration_path`: Path to the zone configuration JSON file.
+  - `--source_video_path`: Path to the source video file.
+  - `--weights`: Path to the model weights file. Default is `'yolov8s.pt'`.
+  - `--device`: Computation device (`'cpu'`, `'mps'` or `'cuda'`). Default is `'cpu'`.
+  - `--classes`: List of class IDs to track. If empty, all classes are tracked.
+  - `--confidence_threshold`: Confidence level for detections (`0` to `1`). Default is `0.3`.
+  - `--iou_threshold`: IOU threshold for non-max suppression. Default is `0.7`.
+
+```bash
+python ultralytics_file_example.py \
+--zone_configuration_path "data/checkout/config.json" \
+--source_video_path "data/checkout/video.mp4" \
+--weights "yolov8x.pt" \
+--device "cpu" \
+--classes 0 \
+--confidence_threshold 0.3 \
+--iou_threshold 0.7
+```
+
+```bash
+python ultralytics_file_example.py \
+--zone_configuration_path "data/traffic/config.json" \
+--source_video_path "data/traffic/video.mp4" \
+--weights "yolov8x.pt" \
+--device "cpu" \
+--classes 2 5 6 7 \
+--confidence_threshold 0.3 \
+--iou_threshold 0.7
+```
+
+### `ultralytics_stream_example`
+
+Script to run object detection on a video stream using the Ultralytics YOLOv8 model.
+
+  - `--zone_configuration_path`: Path to the zone configuration JSON file.
+  - `--rtsp_url`: Complete RTSP URL for the video stream.
+  - `--weights`: Path to the model weights file. Default is `'yolov8s.pt'`.
+  - `--device`: Computation device (`'cpu'`, `'mps'` or `'cuda'`). Default is `'cpu'`.
+  - `--classes`: List of class IDs to track. If empty, all classes are tracked.
+  - `--confidence_threshold`: Confidence level for detections (`0` to `1`). Default is `0.3`.
+  - `--iou_threshold`: IOU threshold for non-max suppression. Default is `0.7`.
+
+```bash
+python ultralytics_stream_example.py \
+--zone_configuration_path "data/checkout/config.json" \
+--rtsp_url "rtsp://localhost:8554/live0.stream" \
+--weights "yolov8x.pt" \
+--device "cpu" \
+--classes 0 \
+--confidence_threshold 0.3 \
+--iou_threshold 0.7
+```
+
+```bash
+python ultralytics_stream_example.py \
+--zone_configuration_path "data/traffic/config.json" \
+--rtsp_url "rtsp://localhost:8554/live0.stream" \
+--weights "yolov8x.pt" \
+--device "cpu" \
+--classes 2 5 6 7 \
+--confidence_threshold 0.3 \
+--iou_threshold 0.7
+```
+
+</details>
+
+## © license
+
+This demo integrates two main components, each with its own licensing:
+
+- ultralytics: The object detection model used in this demo, YOLOv8, is distributed
+  under the [AGPL-3.0 license](https://github.com/ultralytics/ultralytics/blob/main/LICENSE).
+  You can find more details about this license here.
+
+- supervision: The analytics code that powers the zone-based analysis in this demo is
+  based on the Supervision library, which is licensed under the
+  [MIT license](https://github.com/roboflow/supervision/blob/develop/LICENSE.md). This
+  makes the Supervision part of the code fully open source and freely usable in your
+  projects.
@@ -0,0 +1,128 @@
+import argparse
+from typing import List
+
+import cv2
+import numpy as np
+from inference import get_model
+from utils.general import find_in_list, load_zones_config
+from utils.timers import FPSBasedTimer
+
+import supervision as sv
+
+COLORS = sv.ColorPalette.from_hex(["#E6194B", "#3CB44B", "#FFE119", "#3C76D1"])
+COLOR_ANNOTATOR = sv.ColorAnnotator(color=COLORS)
+LABEL_ANNOTATOR = sv.LabelAnnotator(
+    color=COLORS, text_color=sv.Color.from_hex("#000000")
+)
+
+
+def main(
+    source_video_path: str,
+    zone_configuration_path: str,
+    model_id: str,
+    confidence: float,
+    iou: float,
+    classes: List[int],
+) -> None:
+    model = get_model(model_id=model_id)
+    tracker = sv.ByteTrack(minimum_matching_threshold=0.5)
+    video_info = sv.VideoInfo.from_video_path(video_path=source_video_path)
+    frames_generator = sv.get_video_frames_generator(source_video_path)
+
+    polygons = load_zones_config(file_path=zone_configuration_path)
+    zones = [
+        sv.PolygonZone(
+            polygon=polygon,
+            triggering_anchors=(sv.Position.CENTER,),
+        )
+        for polygon in polygons
+    ]
+    timers = [FPSBasedTimer(video_info.fps) for _ in zones]
+
+    for frame in frames_generator:
+        results = model.infer(frame, confidence=confidence, iou_threshold=iou)[0]
+        detections = sv.Detections.from_inference(results)
+        detections = detections[find_in_list(detections.class_id, classes)]
+        detections = tracker.update_with_detections(detections)
+
+        annotated_frame = frame.copy()
+
+        for idx, zone in enumerate(zones):
+            annotated_frame = sv.draw_polygon(
+                scene=annotated_frame, polygon=zone.polygon, color=COLORS.by_idx(idx)
+            )
+
+            detections_in_zone = detections[zone.trigger(detections)]
+            time_in_zone = timers[idx].tick(detections_in_zone)
+            custom_color_lookup = np.full(detections_in_zone.class_id.shape, idx)
+
+            annotated_frame = COLOR_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                custom_color_lookup=custom_color_lookup,
+            )
+            labels = [
+                f"#{tracker_id} {int(time // 60):02d}:{int(time % 60):02d}"
+                for tracker_id, time in zip(detections_in_zone.tracker_id, time_in_zone)
+            ]
+            annotated_frame = LABEL_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                labels=labels,
+                custom_color_lookup=custom_color_lookup,
+            )
+
+        cv2.imshow("Processed Video", annotated_frame)
+        if cv2.waitKey(1) & 0xFF == ord("q"):
+            break
+    cv2.destroyAllWindows()
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Calculating detections dwell time in zones, using video file."
+    )
+    parser.add_argument(
+        "--zone_configuration_path",
+        type=str,
+        required=True,
+        help="Path to the zone configuration JSON file.",
+    )
+    parser.add_argument(
+        "--source_video_path",
+        type=str,
+        required=True,
+        help="Path to the source video file.",
+    )
+    parser.add_argument(
+        "--model_id", type=str, default="yolov8s-640", help="Roboflow model ID."
+    )
+    parser.add_argument(
+        "--confidence_threshold",
+        type=float,
+        default=0.3,
+        help="Confidence level for detections (0 to 1). Default is 0.3.",
+    )
+    parser.add_argument(
+        "--iou_threshold",
+        default=0.7,
+        type=float,
+        help="IOU threshold for non-max suppression. Default is 0.7.",
+    )
+    parser.add_argument(
+        "--classes",
+        nargs="*",
+        type=int,
+        default=[],
+        help="List of class IDs to track. If empty, all classes are tracked.",
+    )
+    args = parser.parse_args()
+
+    main(
+        source_video_path=args.source_video_path,
+        zone_configuration_path=args.zone_configuration_path,
+        model_id=args.model_id,
+        confidence=args.confidence_threshold,
+        iou=args.iou_threshold,
+        classes=args.classes,
+    )
@@ -0,0 +1,138 @@
+import argparse
+from typing import List
+
+import cv2
+import numpy as np
+from inference import get_model
+from utils.general import find_in_list, get_stream_frames_generator, load_zones_config
+from utils.timers import ClockBasedTimer
+
+import supervision as sv
+
+COLORS = sv.ColorPalette.from_hex(["#E6194B", "#3CB44B", "#FFE119", "#3C76D1"])
+COLOR_ANNOTATOR = sv.ColorAnnotator(color=COLORS)
+LABEL_ANNOTATOR = sv.LabelAnnotator(
+    color=COLORS, text_color=sv.Color.from_hex("#000000")
+)
+
+
+def main(
+    rtsp_url: str,
+    zone_configuration_path: str,
+    model_id: str,
+    confidence: float,
+    iou: float,
+    classes: List[int],
+) -> None:
+    model = get_model(model_id=model_id)
+    tracker = sv.ByteTrack(minimum_matching_threshold=0.5)
+    frames_generator = get_stream_frames_generator(rtsp_url=rtsp_url)
+    fps_monitor = sv.FPSMonitor()
+
+    polygons = load_zones_config(file_path=zone_configuration_path)
+    zones = [
+        sv.PolygonZone(
+            polygon=polygon,
+            triggering_anchors=(sv.Position.CENTER,),
+        )
+        for polygon in polygons
+    ]
+    timers = [ClockBasedTimer() for _ in zones]
+
+    for frame in frames_generator:
+        fps_monitor.tick()
+        fps = fps_monitor.fps
+
+        results = model.infer(frame, confidence=confidence, iou_threshold=iou)[0]
+        detections = sv.Detections.from_inference(results)
+        detections = detections[find_in_list(detections.class_id, classes)]
+        detections = tracker.update_with_detections(detections)
+
+        annotated_frame = frame.copy()
+        annotated_frame = sv.draw_text(
+            scene=annotated_frame,
+            text=f"{fps:.1f}",
+            text_anchor=sv.Point(40, 30),
+            background_color=sv.Color.from_hex("#A351FB"),
+            text_color=sv.Color.from_hex("#000000"),
+        )
+
+        for idx, zone in enumerate(zones):
+            annotated_frame = sv.draw_polygon(
+                scene=annotated_frame, polygon=zone.polygon, color=COLORS.by_idx(idx)
+            )
+
+            detections_in_zone = detections[zone.trigger(detections)]
+            time_in_zone = timers[idx].tick(detections_in_zone)
+            custom_color_lookup = np.full(detections_in_zone.class_id.shape, idx)
+
+            annotated_frame = COLOR_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                custom_color_lookup=custom_color_lookup,
+            )
+            labels = [
+                f"#{tracker_id} {int(time // 60):02d}:{int(time % 60):02d}"
+                for tracker_id, time in zip(detections_in_zone.tracker_id, time_in_zone)
+            ]
+            annotated_frame = LABEL_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                labels=labels,
+                custom_color_lookup=custom_color_lookup,
+            )
+
+        cv2.imshow("Processed Video", annotated_frame)
+        if cv2.waitKey(1) & 0xFF == ord("q"):
+            break
+    cv2.destroyAllWindows()
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Calculating detections dwell time in zones, using RTSP stream."
+    )
+    parser.add_argument(
+        "--zone_configuration_path",
+        type=str,
+        required=True,
+        help="Path to the zone configuration JSON file.",
+    )
+    parser.add_argument(
+        "--rtsp_url",
+        type=str,
+        required=True,
+        help="Complete RTSP URL for the video stream.",
+    )
+    parser.add_argument(
+        "--model_id", type=str, default="yolov8s-640", help="Roboflow model ID."
+    )
+    parser.add_argument(
+        "--confidence_threshold",
+        type=float,
+        default=0.3,
+        help="Confidence level for detections (0 to 1). Default is 0.3.",
+    )
+    parser.add_argument(
+        "--iou_threshold",
+        default=0.7,
+        type=float,
+        help="IOU threshold for non-max suppression. Default is 0.7.",
+    )
+    parser.add_argument(
+        "--classes",
+        nargs="*",
+        type=int,
+        default=[],
+        help="List of class IDs to track. If empty, all classes are tracked.",
+    )
+    args = parser.parse_args()
+
+    main(
+        rtsp_url=args.rtsp_url,
+        zone_configuration_path=args.zone_configuration_path,
+        model_id=args.model_id,
+        confidence=args.confidence_threshold,
+        iou=args.iou_threshold,
+        classes=args.classes,
+    )
@@ -0,0 +1,153 @@
+import argparse
+from typing import List
+
+import cv2
+import numpy as np
+from inference import InferencePipeline
+from inference.core.interfaces.camera.entities import VideoFrame
+from utils.general import find_in_list, load_zones_config
+from utils.timers import ClockBasedTimer
+
+import supervision as sv
+
+COLORS = sv.ColorPalette.from_hex(["#E6194B", "#3CB44B", "#FFE119", "#3C76D1"])
+COLOR_ANNOTATOR = sv.ColorAnnotator(color=COLORS)
+LABEL_ANNOTATOR = sv.LabelAnnotator(
+    color=COLORS, text_color=sv.Color.from_hex("#000000")
+)
+
+
+class CustomSink:
+    def __init__(self, zone_configuration_path: str, classes: List[int]):
+        self.classes = classes
+        self.tracker = sv.ByteTrack(minimum_matching_threshold=0.5)
+        self.fps_monitor = sv.FPSMonitor()
+        self.polygons = load_zones_config(file_path=zone_configuration_path)
+        self.timers = [ClockBasedTimer() for _ in self.polygons]
+        self.zones = [
+            sv.PolygonZone(
+                polygon=polygon,
+                triggering_anchors=(sv.Position.CENTER,),
+            )
+            for polygon in self.polygons
+        ]
+
+    def on_prediction(self, result: dict, frame: VideoFrame) -> None:
+        self.fps_monitor.tick()
+        fps = self.fps_monitor.fps
+
+        detections = sv.Detections.from_inference(result)
+        detections = detections[find_in_list(detections.class_id, self.classes)]
+        detections = self.tracker.update_with_detections(detections)
+
+        annotated_frame = frame.image.copy()
+        annotated_frame = sv.draw_text(
+            scene=annotated_frame,
+            text=f"{fps:.1f}",
+            text_anchor=sv.Point(40, 30),
+            background_color=sv.Color.from_hex("#A351FB"),
+            text_color=sv.Color.from_hex("#000000"),
+        )
+
+        for idx, zone in enumerate(self.zones):
+            annotated_frame = sv.draw_polygon(
+                scene=annotated_frame, polygon=zone.polygon, color=COLORS.by_idx(idx)
+            )
+
+            detections_in_zone = detections[zone.trigger(detections)]
+            time_in_zone = self.timers[idx].tick(detections_in_zone)
+            custom_color_lookup = np.full(detections_in_zone.class_id.shape, idx)
+
+            annotated_frame = COLOR_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                custom_color_lookup=custom_color_lookup,
+            )
+            labels = [
+                f"#{tracker_id} {int(time // 60):02d}:{int(time % 60):02d}"
+                for tracker_id, time in zip(detections_in_zone.tracker_id, time_in_zone)
+            ]
+            annotated_frame = LABEL_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                labels=labels,
+                custom_color_lookup=custom_color_lookup,
+            )
+        cv2.imshow("Processed Video", annotated_frame)
+        cv2.waitKey(1)
+
+
+def main(
+    rtsp_url: str,
+    zone_configuration_path: str,
+    model_id: str,
+    confidence: float,
+    iou: float,
+    classes: List[int],
+) -> None:
+    sink = CustomSink(zone_configuration_path=zone_configuration_path, classes=classes)
+
+    pipeline = InferencePipeline.init(
+        model_id=model_id,
+        video_reference=rtsp_url,
+        on_prediction=sink.on_prediction,
+        confidence=confidence,
+        iou_threshold=iou,
+    )
+
+    pipeline.start()
+
+    try:
+        pipeline.join()
+    except KeyboardInterrupt:
+        pipeline.terminate()
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Calculating detections dwell time in zones, using RTSP stream."
+    )
+    parser.add_argument(
+        "--zone_configuration_path",
+        type=str,
+        required=True,
+        help="Path to the zone configuration JSON file.",
+    )
+    parser.add_argument(
+        "--rtsp_url",
+        type=str,
+        required=True,
+        help="Complete RTSP URL for the video stream.",
+    )
+    parser.add_argument(
+        "--model_id", type=str, default="yolov8s-640", help="Roboflow model ID."
+    )
+    parser.add_argument(
+        "--confidence_threshold",
+        type=float,
+        default=0.3,
+        help="Confidence level for detections (0 to 1). Default is 0.3.",
+    )
+    parser.add_argument(
+        "--iou_threshold",
+        default=0.7,
+        type=float,
+        help="IOU threshold for non-max suppression. Default is 0.7.",
+    )
+    parser.add_argument(
+        "--classes",
+        nargs="*",
+        type=int,
+        default=[],
+        help="List of class IDs to track. If empty, all classes are tracked.",
+    )
+    args = parser.parse_args()
+
+    main(
+        rtsp_url=args.rtsp_url,
+        zone_configuration_path=args.zone_configuration_path,
+        model_id=args.model_id,
+        confidence=args.confidence_threshold,
+        iou=args.iou_threshold,
+        classes=args.classes,
+    )
@@ -0,0 +1,5 @@
+opencv-python
+supervision>=0.20.0
+ultralytics
+inference==0.9.17
+pytube
@@ -0,0 +1,46 @@
+import argparse
+import os
+from typing import Optional
+
+from pytube import YouTube
+
+
+def main(url: str, output_path: Optional[str], file_name: Optional[str]) -> None:
+    yt = YouTube(url)
+    stream = yt.streams.get_highest_resolution()
+
+    if not os.path.exists(output_path):
+        os.makedirs(output_path)
+
+    stream.download(output_path=output_path, filename=file_name)
+    final_name = file_name if file_name else yt.title
+    final_path = output_path if output_path else "current directory"
+    print(f"Download completed! Video saved as '{final_name}' in '{final_path}'.")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Download a specific YouTube video by providing its URL."
+    )
+    parser.add_argument(
+        "--url",
+        type=str,
+        required=True,
+        help="The full URL of the YouTube video you wish to download.",
+    )
+    parser.add_argument(
+        "--output_path",
+        type=str,
+        default="data/source",
+        required=False,
+        help="Optional. Specifies the directory where the video will be saved.",
+    )
+    parser.add_argument(
+        "--file_name",
+        type=str,
+        default="video.mp4",
+        required=False,
+        help="Optional. Sets the name of the saved video file.",
+    )
+    args = parser.parse_args()
+    main(url=args.url, output_path=args.output_path, file_name=args.file_name)
@@ -0,0 +1,176 @@
+import argparse
+import json
+import os
+from typing import Any, Optional, Tuple
+
+import cv2
+import numpy as np
+
+import supervision as sv
+
+KEY_ENTER = 13
+KEY_NEWLINE = 10
+KEY_ESCAPE = 27
+KEY_QUIT = ord("q")
+KEY_SAVE = ord("s")
+
+THICKNESS = 2
+COLORS = sv.ColorPalette.DEFAULT
+WINDOW_NAME = "Draw Zones"
+POLYGONS = [[]]
+
+current_mouse_position: Optional[Tuple[int, int]] = None
+
+
+def resolve_source(source_path: str) -> Optional[np.ndarray]:
+    if not os.path.exists(source_path):
+        return None
+
+    image = cv2.imread(source_path)
+    if image is not None:
+        return image
+
+    frame_generator = sv.get_video_frames_generator(source_path=source_path)
+    frame = next(frame_generator)
+    return frame
+
+
+def mouse_event(event: int, x: int, y: int, flags: int, param: Any) -> None:
+    global current_mouse_position
+    if event == cv2.EVENT_MOUSEMOVE:
+        current_mouse_position = (x, y)
+    elif event == cv2.EVENT_LBUTTONDOWN:
+        POLYGONS[-1].append((x, y))
+
+
+def redraw(image: np.ndarray, original_image: np.ndarray) -> None:
+    global POLYGONS, current_mouse_position
+    image[:] = original_image.copy()
+    for idx, polygon in enumerate(POLYGONS):
+        color = (
+            COLORS.by_idx(idx).as_bgr()
+            if idx < len(POLYGONS) - 1
+            else sv.Color.WHITE.as_bgr()
+        )
+
+        if len(polygon) > 1:
+            for i in range(1, len(polygon)):
+                cv2.line(
+                    img=image,
+                    pt1=polygon[i - 1],
+                    pt2=polygon[i],
+                    color=color,
+                    thickness=THICKNESS,
+                )
+            if idx < len(POLYGONS) - 1:
+                cv2.line(
+                    img=image,
+                    pt1=polygon[-1],
+                    pt2=polygon[0],
+                    color=color,
+                    thickness=THICKNESS,
+                )
+        if idx == len(POLYGONS) - 1 and current_mouse_position is not None and polygon:
+            cv2.line(
+                img=image,
+                pt1=polygon[-1],
+                pt2=current_mouse_position,
+                color=color,
+                thickness=THICKNESS,
+            )
+    cv2.imshow(WINDOW_NAME, image)
+
+
+def close_and_finalize_polygon(image: np.ndarray, original_image: np.ndarray) -> None:
+    if len(POLYGONS[-1]) > 2:
+        cv2.line(
+            img=image,
+            pt1=POLYGONS[-1][-1],
+            pt2=POLYGONS[-1][0],
+            color=COLORS.by_idx(0).as_bgr(),
+            thickness=THICKNESS,
+        )
+    POLYGONS.append([])
+    image[:] = original_image.copy()
+    redraw_polygons(image)
+    cv2.imshow(WINDOW_NAME, image)
+
+
+def redraw_polygons(image: np.ndarray) -> None:
+    for idx, polygon in enumerate(POLYGONS[:-1]):
+        if len(polygon) > 1:
+            color = COLORS.by_idx(idx).as_bgr()
+            for i in range(len(polygon) - 1):
+                cv2.line(
+                    img=image,
+                    pt1=polygon[i],
+                    pt2=polygon[i + 1],
+                    color=color,
+                    thickness=THICKNESS,
+                )
+            cv2.line(
+                img=image,
+                pt1=polygon[-1],
+                pt2=polygon[0],
+                color=color,
+                thickness=THICKNESS,
+            )
+
+
+def save_polygons_to_json(polygons, target_path):
+    data_to_save = polygons if polygons[-1] else polygons[:-1]
+    with open(target_path, "w") as f:
+        json.dump(data_to_save, f)
+
+
+def main(source_path: str, zone_configuration_path: str) -> None:
+    global current_mouse_position
+    original_image = resolve_source(source_path=source_path)
+    if original_image is None:
+        print("Failed to load source image.")
+        return
+
+    image = original_image.copy()
+    cv2.imshow(WINDOW_NAME, image)
+    cv2.setMouseCallback(WINDOW_NAME, mouse_event, image)
+
+    while True:
+        key = cv2.waitKey(1) & 0xFF
+        if key == KEY_ENTER or key == KEY_NEWLINE:
+            close_and_finalize_polygon(image, original_image)
+        elif key == KEY_ESCAPE:
+            POLYGONS[-1] = []
+            current_mouse_position = None
+        elif key == KEY_SAVE:
+            save_polygons_to_json(POLYGONS, zone_configuration_path)
+            print(f"Polygons saved to {zone_configuration_path}")
+            break
+        redraw(image, original_image)
+        if key == KEY_QUIT:
+            break
+
+    cv2.destroyAllWindows()
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Interactively draw polygons on images or video frames and save "
+        "the annotations."
+    )
+    parser.add_argument(
+        "--source_path",
+        type=str,
+        required=True,
+        help="Path to the source image or video file for drawing polygons.",
+    )
+    parser.add_argument(
+        "--zone_configuration_path",
+        type=str,
+        required=True,
+        help="Path where the polygon annotations will be saved as a JSON file.",
+    )
+    arguments = parser.parse_args()
+    main(
+        source_path=arguments.source_path,
+        zone_configuration_path=arguments.zone_configuration_path,
+    )
@@ -0,0 +1,104 @@
+import argparse
+import os
+import subprocess
+import tempfile
+from glob import glob
+from threading import Thread
+
+import yaml
+
+SERVER_CONFIG = {"protocols": ["tcp"], "paths": {"all": {"source": "publisher"}}}
+BASE_STREAM_URL = "rtsp://localhost:8554/live"
+
+
+def main(video_directory: str, number_of_streams: int) -> None:
+    video_files = find_video_files_in_directory(video_directory, number_of_streams)
+    try:
+        with tempfile.TemporaryDirectory() as temporary_directory:
+            config_file_path = create_server_config_file(temporary_directory)
+            run_rtsp_server(config_path=config_file_path)
+            stream_videos(video_files)
+    finally:
+        stop_rtsp_server()
+
+
+def find_video_files_in_directory(directory: str, limit: int) -> list:
+    video_formats = ["*.mp4", "*.webm"]
+    video_paths = []
+    for video_format in video_formats:
+        video_paths.extend(glob(os.path.join(directory, video_format)))
+    return video_paths[:limit]
+
+
+def create_server_config_file(directory: str) -> str:
+    config_path = os.path.join(directory, "rtsp-simple-server.yml")
+    with open(config_path, "w") as config_file:
+        yaml.dump(SERVER_CONFIG, config_file)
+    return config_path
+
+
+def run_rtsp_server(config_path: str) -> None:
+    command = (
+        "docker run --rm --name rtsp_server -d -v "
+        f"{config_path}:/rtsp-simple-server.yml -p 8554:8554 "
+        "aler9/rtsp-simple-server:v1.3.0"
+    )
+    if run_command(command.split()) != 0:
+        raise RuntimeError("Could not start the RTSP server!")
+
+
+def stop_rtsp_server() -> None:
+    run_command("docker kill rtsp_server".split())
+
+
+def stream_videos(video_files: list) -> None:
+    threads = []
+    for index, video_file in enumerate(video_files):
+        stream_url = f"{BASE_STREAM_URL}{index}.stream"
+        print(f"Streaming {video_file} under {stream_url}")
+        thread = stream_video_to_url(video_file, stream_url)
+        threads.append(thread)
+    for thread in threads:
+        thread.join()
+
+
+def stream_video_to_url(video_path: str, stream_url: str) -> Thread:
+    command = (
+        f"ffmpeg -re -stream_loop -1 -i {video_path} "
+        f"-f rtsp -rtsp_transport tcp {stream_url}"
+    )
+    return run_command_in_thread(command.split())
+
+
+def run_command_in_thread(command: list) -> Thread:
+    thread = Thread(target=run_command, args=(command,))
+    thread.start()
+    return thread
+
+
+def run_command(command: list) -> int:
+    process = subprocess.run(command)
+    return process.returncode
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Script to stream videos using RTSP protocol."
+    )
+    parser.add_argument(
+        "--video_directory",
+        type=str,
+        required=True,
+        help="Directory containing video files to stream.",
+    )
+    parser.add_argument(
+        "--number_of_streams",
+        type=int,
+        default=6,
+        help="Number of video files to stream.",
+    )
+    arguments = parser.parse_args()
+    main(
+        video_directory=arguments.video_directory,
+        number_of_streams=arguments.number_of_streams,
+    )
@@ -0,0 +1,140 @@
+import argparse
+from typing import List
+
+import cv2
+import numpy as np
+from ultralytics import YOLO
+from utils.general import find_in_list, load_zones_config
+from utils.timers import FPSBasedTimer
+
+import supervision as sv
+
+COLORS = sv.ColorPalette.from_hex(["#E6194B", "#3CB44B", "#FFE119", "#3C76D1"])
+COLOR_ANNOTATOR = sv.ColorAnnotator(color=COLORS)
+LABEL_ANNOTATOR = sv.LabelAnnotator(
+    color=COLORS, text_color=sv.Color.from_hex("#000000")
+)
+
+
+def main(
+    source_video_path: str,
+    zone_configuration_path: str,
+    weights: str,
+    device: str,
+    confidence: float,
+    iou: float,
+    classes: List[int],
+) -> None:
+    model = YOLO(weights)
+    tracker = sv.ByteTrack(minimum_matching_threshold=0.5)
+    video_info = sv.VideoInfo.from_video_path(video_path=source_video_path)
+    frames_generator = sv.get_video_frames_generator(source_video_path)
+
+    polygons = load_zones_config(file_path=zone_configuration_path)
+    zones = [
+        sv.PolygonZone(
+            polygon=polygon,
+            triggering_anchors=(sv.Position.CENTER,),
+        )
+        for polygon in polygons
+    ]
+    timers = [FPSBasedTimer(video_info.fps) for _ in zones]
+
+    for frame in frames_generator:
+        results = model(frame, verbose=False, device=device, conf=confidence)[0]
+        detections = sv.Detections.from_ultralytics(results)
+        detections = detections[find_in_list(detections.class_id, classes)]
+        detections = detections.with_nms(threshold=iou)
+        detections = tracker.update_with_detections(detections)
+
+        annotated_frame = frame.copy()
+
+        for idx, zone in enumerate(zones):
+            annotated_frame = sv.draw_polygon(
+                scene=annotated_frame, polygon=zone.polygon, color=COLORS.by_idx(idx)
+            )
+
+            detections_in_zone = detections[zone.trigger(detections)]
+            time_in_zone = timers[idx].tick(detections_in_zone)
+            custom_color_lookup = np.full(detections_in_zone.class_id.shape, idx)
+
+            annotated_frame = COLOR_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                custom_color_lookup=custom_color_lookup,
+            )
+            labels = [
+                f"#{tracker_id} {int(time // 60):02d}:{int(time % 60):02d}"
+                for tracker_id, time in zip(detections_in_zone.tracker_id, time_in_zone)
+            ]
+            annotated_frame = LABEL_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                labels=labels,
+                custom_color_lookup=custom_color_lookup,
+            )
+
+        cv2.imshow("Processed Video", annotated_frame)
+        if cv2.waitKey(1) & 0xFF == ord("q"):
+            break
+    cv2.destroyAllWindows()
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Calculating detections dwell time in zones, using video file."
+    )
+    parser.add_argument(
+        "--zone_configuration_path",
+        type=str,
+        required=True,
+        help="Path to the zone configuration JSON file.",
+    )
+    parser.add_argument(
+        "--source_video_path",
+        type=str,
+        required=True,
+        help="Path to the source video file.",
+    )
+    parser.add_argument(
+        "--weights",
+        type=str,
+        default="yolov8s.pt",
+        help="Path to the model weights file. Default is 'yolov8s.pt'.",
+    )
+    parser.add_argument(
+        "--device",
+        type=str,
+        default="cpu",
+        help="Computation device ('cpu', 'mps' or 'cuda'). Default is 'cpu'.",
+    )
+    parser.add_argument(
+        "--confidence_threshold",
+        type=float,
+        default=0.3,
+        help="Confidence level for detections (0 to 1). Default is 0.3.",
+    )
+    parser.add_argument(
+        "--iou_threshold",
+        default=0.7,
+        type=float,
+        help="IOU threshold for non-max suppression. Default is 0.7.",
+    )
+    parser.add_argument(
+        "--classes",
+        nargs="*",
+        type=int,
+        default=[],
+        help="List of class IDs to track. If empty, all classes are tracked.",
+    )
+    args = parser.parse_args()
+
+    main(
+        source_video_path=args.source_video_path,
+        zone_configuration_path=args.zone_configuration_path,
+        weights=args.weights,
+        device=args.device,
+        confidence=args.confidence_threshold,
+        iou=args.iou_threshold,
+        classes=args.classes,
+    )
@@ -0,0 +1,150 @@
+import argparse
+from typing import List
+
+import cv2
+import numpy as np
+from ultralytics import YOLO
+from utils.general import find_in_list, get_stream_frames_generator, load_zones_config
+from utils.timers import ClockBasedTimer
+
+import supervision as sv
+
+COLORS = sv.ColorPalette.from_hex(["#E6194B", "#3CB44B", "#FFE119", "#3C76D1"])
+COLOR_ANNOTATOR = sv.ColorAnnotator(color=COLORS)
+LABEL_ANNOTATOR = sv.LabelAnnotator(
+    color=COLORS, text_color=sv.Color.from_hex("#000000")
+)
+
+
+def main(
+    rtsp_url: str,
+    zone_configuration_path: str,
+    weights: str,
+    device: str,
+    confidence: float,
+    iou: float,
+    classes: List[int],
+) -> None:
+    model = YOLO(weights)
+    tracker = sv.ByteTrack(minimum_matching_threshold=0.5)
+    frames_generator = get_stream_frames_generator(rtsp_url=rtsp_url)
+    fps_monitor = sv.FPSMonitor()
+
+    polygons = load_zones_config(file_path=zone_configuration_path)
+    zones = [
+        sv.PolygonZone(
+            polygon=polygon,
+            triggering_anchors=(sv.Position.CENTER,),
+        )
+        for polygon in polygons
+    ]
+    timers = [ClockBasedTimer() for _ in zones]
+
+    for frame in frames_generator:
+        fps_monitor.tick()
+        fps = fps_monitor.fps
+
+        results = model(frame, verbose=False, device=device, conf=confidence)[0]
+        detections = sv.Detections.from_ultralytics(results)
+        detections = detections[find_in_list(detections.class_id, classes)]
+        detections = detections.with_nms(threshold=iou)
+        detections = tracker.update_with_detections(detections)
+
+        annotated_frame = frame.copy()
+        annotated_frame = sv.draw_text(
+            scene=annotated_frame,
+            text=f"{fps:.1f}",
+            text_anchor=sv.Point(40, 30),
+            background_color=sv.Color.from_hex("#A351FB"),
+            text_color=sv.Color.from_hex("#000000"),
+        )
+
+        for idx, zone in enumerate(zones):
+            annotated_frame = sv.draw_polygon(
+                scene=annotated_frame, polygon=zone.polygon, color=COLORS.by_idx(idx)
+            )
+
+            detections_in_zone = detections[zone.trigger(detections)]
+            time_in_zone = timers[idx].tick(detections_in_zone)
+            custom_color_lookup = np.full(detections_in_zone.class_id.shape, idx)
+
+            annotated_frame = COLOR_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                custom_color_lookup=custom_color_lookup,
+            )
+            labels = [
+                f"#{tracker_id} {int(time // 60):02d}:{int(time % 60):02d}"
+                for tracker_id, time in zip(detections_in_zone.tracker_id, time_in_zone)
+            ]
+            annotated_frame = LABEL_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                labels=labels,
+                custom_color_lookup=custom_color_lookup,
+            )
+
+        cv2.imshow("Processed Video", annotated_frame)
+        if cv2.waitKey(1) & 0xFF == ord("q"):
+            break
+    cv2.destroyAllWindows()
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Calculating detections dwell time in zones, using RTSP stream."
+    )
+    parser.add_argument(
+        "--zone_configuration_path",
+        type=str,
+        required=True,
+        help="Path to the zone configuration JSON file.",
+    )
+    parser.add_argument(
+        "--rtsp_url",
+        type=str,
+        required=True,
+        help="Complete RTSP URL for the video stream.",
+    )
+    parser.add_argument(
+        "--weights",
+        type=str,
+        default="yolov8s.pt",
+        help="Path to the model weights file. Default is 'yolov8s.pt'.",
+    )
+    parser.add_argument(
+        "--device",
+        type=str,
+        default="cpu",
+        help="Computation device ('cpu', 'mps' or 'cuda'). Default is 'cpu'.",
+    )
+    parser.add_argument(
+        "--confidence_threshold",
+        type=float,
+        default=0.3,
+        help="Confidence level for detections (0 to 1). Default is 0.3.",
+    )
+    parser.add_argument(
+        "--iou_threshold",
+        default=0.7,
+        type=float,
+        help="IOU threshold for non-max suppression. Default is 0.7.",
+    )
+    parser.add_argument(
+        "--classes",
+        nargs="*",
+        type=int,
+        default=[],
+        help="List of class IDs to track. If empty, all classes are tracked.",
+    )
+    args = parser.parse_args()
+
+    main(
+        rtsp_url=args.rtsp_url,
+        zone_configuration_path=args.zone_configuration_path,
+        weights=args.weights,
+        device=args.device,
+        confidence=args.confidence_threshold,
+        iou=args.iou_threshold,
+        classes=args.classes,
+    )
@@ -0,0 +1,168 @@
+import argparse
+from typing import List
+
+import cv2
+import numpy as np
+from inference import InferencePipeline
+from inference.core.interfaces.camera.entities import VideoFrame
+from ultralytics import YOLO
+from utils.general import find_in_list, load_zones_config
+from utils.timers import ClockBasedTimer
+
+import supervision as sv
+
+COLORS = sv.ColorPalette.from_hex(["#E6194B", "#3CB44B", "#FFE119", "#3C76D1"])
+COLOR_ANNOTATOR = sv.ColorAnnotator(color=COLORS)
+LABEL_ANNOTATOR = sv.LabelAnnotator(
+    color=COLORS, text_color=sv.Color.from_hex("#000000")
+)
+
+
+class CustomSink:
+    def __init__(self, zone_configuration_path: str, classes: List[int]):
+        self.classes = classes
+        self.tracker = sv.ByteTrack(minimum_matching_threshold=0.8)
+        self.fps_monitor = sv.FPSMonitor()
+        self.polygons = load_zones_config(file_path=zone_configuration_path)
+        self.timers = [ClockBasedTimer() for _ in self.polygons]
+        self.zones = [
+            sv.PolygonZone(
+                polygon=polygon,
+                triggering_anchors=(sv.Position.CENTER,),
+            )
+            for polygon in self.polygons
+        ]
+
+    def on_prediction(self, detections: sv.Detections, frame: VideoFrame) -> None:
+        self.fps_monitor.tick()
+        fps = self.fps_monitor.fps
+
+        detections = detections[find_in_list(detections.class_id, self.classes)]
+        detections = self.tracker.update_with_detections(detections)
+
+        annotated_frame = frame.image.copy()
+        annotated_frame = sv.draw_text(
+            scene=annotated_frame,
+            text=f"{fps:.1f}",
+            text_anchor=sv.Point(40, 30),
+            background_color=sv.Color.from_hex("#A351FB"),
+            text_color=sv.Color.from_hex("#000000"),
+        )
+
+        for idx, zone in enumerate(self.zones):
+            annotated_frame = sv.draw_polygon(
+                scene=annotated_frame, polygon=zone.polygon, color=COLORS.by_idx(idx)
+            )
+
+            detections_in_zone = detections[zone.trigger(detections)]
+            time_in_zone = self.timers[idx].tick(detections_in_zone)
+            custom_color_lookup = np.full(detections_in_zone.class_id.shape, idx)
+
+            annotated_frame = COLOR_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                custom_color_lookup=custom_color_lookup,
+            )
+            labels = [
+                f"#{tracker_id} {int(time // 60):02d}:{int(time % 60):02d}"
+                for tracker_id, time in zip(detections_in_zone.tracker_id, time_in_zone)
+            ]
+            annotated_frame = LABEL_ANNOTATOR.annotate(
+                scene=annotated_frame,
+                detections=detections_in_zone,
+                labels=labels,
+                custom_color_lookup=custom_color_lookup,
+            )
+        cv2.imshow("Processed Video", annotated_frame)
+        cv2.waitKey(1)
+
+
+def main(
+    rtsp_url: str,
+    zone_configuration_path: str,
+    weights: str,
+    device: str,
+    confidence: float,
+    iou: float,
+    classes: List[int],
+) -> None:
+    model = YOLO(weights)
+
+    def inference_callback(frame: VideoFrame) -> sv.Detections:
+        results = model(frame.image, verbose=False, conf=confidence, device=device)[0]
+        return sv.Detections.from_ultralytics(results).with_nms(threshold=iou)
+
+    sink = CustomSink(zone_configuration_path=zone_configuration_path, classes=classes)
+
+    pipeline = InferencePipeline.init_with_custom_logic(
+        video_reference=rtsp_url,
+        on_video_frame=inference_callback,
+        on_prediction=sink.on_prediction,
+    )
+
+    pipeline.start()
+
+    try:
+        pipeline.join()
+    except KeyboardInterrupt:
+        pipeline.terminate()
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Calculating detections dwell time in zones, using RTSP stream."
+    )
+    parser.add_argument(
+        "--zone_configuration_path",
+        type=str,
+        required=True,
+        help="Path to the zone configuration JSON file.",
+    )
+    parser.add_argument(
+        "--rtsp_url",
+        type=str,
+        required=True,
+        help="Complete RTSP URL for the video stream.",
+    )
+    parser.add_argument(
+        "--weights",
+        type=str,
+        default="yolov8s.pt",
+        help="Path to the model weights file. Default is 'yolov8s.pt'.",
+    )
+    parser.add_argument(
+        "--device",
+        type=str,
+        default="cpu",
+        help="Computation device ('cpu', 'mps' or 'cuda'). Default is 'cpu'.",
+    )
+    parser.add_argument(
+        "--confidence_threshold",
+        type=float,
+        default=0.3,
+        help="Confidence level for detections (0 to 1). Default is 0.3.",
+    )
+    parser.add_argument(
+        "--iou_threshold",
+        default=0.7,
+        type=float,
+        help="IOU threshold for non-max suppression. Default is 0.7.",
+    )
+    parser.add_argument(
+        "--classes",
+        nargs="*",
+        type=int,
+        default=[],
+        help="List of class IDs to track. If empty, all classes are tracked.",
+    )
+    args = parser.parse_args()
+
+    main(
+        rtsp_url=args.rtsp_url,
+        zone_configuration_path=args.zone_configuration_path,
+        weights=args.weights,
+        device=args.device,
+        confidence=args.confidence_threshold,
+        iou=args.iou_threshold,
+        classes=args.classes,
+    )
@@ -0,0 +1,66 @@
+import json
+from typing import Generator, List
+
+import cv2
+import numpy as np
+
+
+def load_zones_config(file_path: str) -> List[np.ndarray]:
+    """
+    Load polygon zone configurations from a JSON file.
+
+    This function reads a JSON file which contains polygon coordinates, and
+    converts them into a list of NumPy arrays. Each polygon is represented as
+    a NumPy array of coordinates.
+
+    Args:
+        file_path (str): The path to the JSON configuration file.
+
+    Returns:
+        List[np.ndarray]: A list of polygons, each represented as a NumPy array.
+    """
+    with open(file_path, "r") as file:
+        data = json.load(file)
+        return [np.array(polygon, np.int32) for polygon in data]
+
+
+def find_in_list(array: np.ndarray, search_list: List[int]) -> np.ndarray:
+    """Determines if elements of a numpy array are present in a list.
+
+    Args:
+        array (np.ndarray): The numpy array of integers to check.
+        search_list (List[int]): The list of integers to search within.
+
+    Returns:
+        np.ndarray: A numpy array of booleans, where each boolean indicates whether
+        the corresponding element in `array` is found in `search_list`.
+    """
+    if not search_list:
+        return np.ones(array.shape, dtype=bool)
+    else:
+        return np.isin(array, search_list)
+
+
+def get_stream_frames_generator(rtsp_url: str) -> Generator[np.ndarray, None, None]:
+    """
+    Generator function to yield frames from an RTSP stream.
+
+    Args:
+        rtsp_url (str): URL of the RTSP video stream.
+
+    Yields:
+        np.ndarray: The next frame from the video stream.
+    """
+    cap = cv2.VideoCapture(rtsp_url)
+    if not cap.isOpened():
+        raise Exception("Error: Could not open video stream.")
+
+    try:
+        while True:
+            ret, frame = cap.read()
+            if not ret:
+                print("End of stream or error reading frame.")
+                break
+            yield frame
+    finally:
+        cap.release()
@@ -0,0 +1,88 @@
+from datetime import datetime
+from typing import Dict
+
+import numpy as np
+
+import supervision as sv
+
+
+class FPSBasedTimer:
+    """
+    A timer that calculates the duration each object has been detected based on frames
+    per second (FPS).
+
+    Attributes:
+        fps (int): The frame rate of the video stream, used to calculate time durations.
+        frame_id (int): The current frame number in the sequence.
+        tracker_id2frame_id (Dict[int, int]): Maps each tracker's ID to the frame number
+            at which it was first detected.
+    """
+
+    def __init__(self, fps: int = 30) -> None:
+        """Initializes the FPSBasedTimer with the specified frames per second rate.
+
+        Args:
+            fps (int, optional): The frame rate of the video stream. Defaults to 30.
+        """
+        self.fps = fps
+        self.frame_id = 0
+        self.tracker_id2frame_id: Dict[int, int] = {}
+
+    def tick(self, detections: sv.Detections) -> np.ndarray:
+        """Processes the current frame, updating time durations for each tracker.
+
+        Args:
+            detections: The detections for the current frame, including tracker IDs.
+
+        Returns:
+            np.ndarray: Time durations (in seconds) for each detected tracker, since
+                their first detection.
+        """
+        self.frame_id += 1
+        times = []
+
+        for tracker_id in detections.tracker_id:
+            self.tracker_id2frame_id.setdefault(tracker_id, self.frame_id)
+
+            start_frame_id = self.tracker_id2frame_id[tracker_id]
+            time_duration = (self.frame_id - start_frame_id) / self.fps
+            times.append(time_duration)
+
+        return np.array(times)
+
+
+class ClockBasedTimer:
+    """
+    A timer that calculates the duration each object has been detected based on the
+    system clock.
+
+    Attributes:
+        tracker_id2start_time (Dict[int, datetime]): Maps each tracker's ID to the
+            datetime when it was first detected.
+    """
+
+    def __init__(self) -> None:
+        """Initializes the ClockBasedTimer."""
+        self.tracker_id2start_time: Dict[int, datetime] = {}
+
+    def tick(self, detections: sv.Detections) -> np.ndarray:
+        """Processes the current frame, updating time durations for each tracker.
+
+        Args:
+            detections: The detections for the current frame, including tracker IDs.
+
+        Returns:
+            np.ndarray: Time durations (in seconds) for each detected tracker, since
+                their first detection.
+        """
+        current_time = datetime.now()
+        times = []
+
+        for tracker_id in detections.tracker_id:
+            self.tracker_id2start_time.setdefault(tracker_id, current_time)
+
+            start_time = self.tracker_id2start_time[tracker_id]
+            time_duration = (current_time - start_time).total_seconds()
+            times.append(time_duration)
+
+        return np.array(times)
@@ -1,4 +1,4 @@
-inference
-supervision
+inference==0.9.17
+supervision==0.19.0
 tqdm
 ultralytics
@@ -1,6 +1,6 @@
 import argparse
 import os
-from typing import Dict, Iterable, List, Set, Tuple
+from typing import Dict, Iterable, List, Set

 import cv2
 import numpy as np
@@ -60,13 +60,11 @@ class DetectionsManager:

 def initiate_polygon_zones(
    polygons: List[np.ndarray],
-    frame_resolution_wh: Tuple[int, int],
    triggering_anchors: Iterable[sv.Position] = [sv.Position.CENTER],
 ) -> List[sv.PolygonZone]:
    return [
        sv.PolygonZone(
            polygon=polygon,
-            frame_resolution_wh=frame_resolution_wh,
            triggering_anchors=triggering_anchors,
        )
        for polygon in polygons
@@ -92,12 +90,8 @@ class VideoProcessor:
        self.tracker = sv.ByteTrack()

        self.video_info = sv.VideoInfo.from_video_path(source_video_path)
-        self.zones_in = initiate_polygon_zones(
-            ZONE_IN_POLYGONS, self.video_info.resolution_wh, [sv.Position.CENTER]
-        )
-        self.zones_out = initiate_polygon_zones(
-            ZONE_OUT_POLYGONS, self.video_info.resolution_wh, [sv.Position.CENTER]
-        )
+        self.zones_in = initiate_polygon_zones(ZONE_IN_POLYGONS, [sv.Position.CENTER])
+        self.zones_out = initiate_polygon_zones(ZONE_OUT_POLYGONS, [sv.Position.CENTER])

        self.bounding_box_annotator = sv.BoundingBoxAnnotator(color=COLORS)
        self.label_annotator = sv.LabelAnnotator(
@@ -1,5 +1,5 @@
 gdown
-inference
-supervision>=0.19.0rc5
+inference==0.9.17
+supervision>=0.20.0
 tqdm
 ultralytics
@@ -1,5 +1,5 @@
 import argparse
-from typing import Dict, Iterable, List, Set, Tuple
+from typing import Dict, Iterable, List, Set

 import cv2
 import numpy as np
@@ -58,13 +58,11 @@ class DetectionsManager:

 def initiate_polygon_zones(
    polygons: List[np.ndarray],
-    frame_resolution_wh: Tuple[int, int],
    triggering_anchors: Iterable[sv.Position] = [sv.Position.CENTER],
 ) -> List[sv.PolygonZone]:
    return [
        sv.PolygonZone(
            polygon=polygon,
-            frame_resolution_wh=frame_resolution_wh,
            triggering_anchors=triggering_anchors,
        )
        for polygon in polygons
@@ -89,12 +87,8 @@ class VideoProcessor:
        self.tracker = sv.ByteTrack()

        self.video_info = sv.VideoInfo.from_video_path(source_video_path)
-        self.zones_in = initiate_polygon_zones(
-            ZONE_IN_POLYGONS, self.video_info.resolution_wh, [sv.Position.CENTER]
-        )
-        self.zones_out = initiate_polygon_zones(
-            ZONE_OUT_POLYGONS, self.video_info.resolution_wh, [sv.Position.CENTER]
-        )
+        self.zones_in = initiate_polygon_zones(ZONE_IN_POLYGONS, [sv.Position.CENTER])
+        self.zones_out = initiate_polygon_zones(ZONE_OUT_POLYGONS, [sv.Position.CENTER])

        self.bounding_box_annotator = sv.BoundingBoxAnnotator(color=COLORS)
        self.label_annotator = sv.LabelAnnotator(
@@ -35,18 +35,26 @@ extra_css:


 nav:
-  - Home: index.md
-  - How to:
+  - Supervision: index.md
+  - Learn:
      - Detect and Annotate: how_to/detect_and_annotate.md
-      - Track Objects: how_to/track_objects.md
+      - Save Detections: how_to/save_detections.md
      - Filter Detections: how_to/filter_detections.md
-  - API:
-    - Annotators: annotators.md
-    - Classifications:
-        - Core: classification/core.md
-    - Detections:
+      - Detect Small Objects: how_to/detect_small_objects.md
+      - Track Objects on Video: how_to/track_objects.md
+
+  - Reference - Code API:
+    - Detection and Segmentation:
        - Core: detection/core.md
+        - Annotators: detection/annotators.md
+        - Metrics: detection/metrics.md
+        - Double Detection Filter: detection/double_detection_filter.md
        - Utils: detection/utils.md
+    - Keypoint Detection:
+        - Core: keypoint/core.md
+        - Annotators: keypoint/annotators.md
+    - Classification:
+        - Core: classification/core.md
    - Tools:
        - Line Zone: detection/tools/line_zone.md
        - Polygon Zone: detection/tools/polygon_zone.md
@@ -54,26 +62,24 @@ nav:
        - Detection Smoother: detection/tools/smoother.md
        - Save Detections: detection/tools/save_detections.md
    - Trackers: trackers.md
-    - Datasets: datasets.md
-    - Metrics:
-      - Object Detection: metrics/detection.md
-    - Draw:
-        - Color: draw/color.md
-        - Utils: draw/utils.md
-    - Geometry:
-        - Position: geometry/core.md
+    - Datasets:
+        - Core: datasets/core.md
+        - Utils: datasets/utils.md
    - Utils:
        - Video: utils/video.md
        - Image: utils/image.md
+        - Iterables: utils/iterables.md
        - Notebook: utils/notebook.md
        - File: utils/file.md
+        - Draw: utils/draw.md
+        - Geometry: utils/geometry.md
    - Assets: assets.md
  - Cookbooks: cookbooks.md
  - Contribute:
    - Contributing: contributing.md
    - Code of Conduct: code_of_conduct.md
    - License: license.md
-  - Changelog:
+  - Release Notes:
        - Changelog: changelog.md
        - Deprecated: deprecated.md

@@ -172,3 +178,12 @@ extra_javascript:
  - "javascripts/init_kapa_widget.js"
  - "javascripts/cookbooks-card.js"
  - "https://cdnjs.cloudflare.com/ajax/libs/dompurify/3.0.8/purify.min.js"
+
+# Messages shown during document build
+# Reference: https://www.mkdocs.org/user-guide/configuration/#validation
+# Values: [warn, info, ignore]
+validation:
+  nav:
+    absolute_links: ignore
+  links:
+    absolute_links: ignore
@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "supervision"
-version = "0.19.0"
+version = "0.22.0rc1"
 description = "A set of easy-to-use utils that will come in handy in any Computer Vision project"
 authors = ["Piotr Skalski <piotr.skalski92@gmail.com>"]
 maintainers = ["Piotr Skalski <piotr.skalski92@gmail.com>"]
@@ -42,8 +42,8 @@ pyyaml = ">=5.3"
 defusedxml = "^0.7.1"
 opencv-python = { version = ">=4.5.5.64", optional = true }
 opencv-python-headless = ">=4.5.5.64"
-requests = { version = ">=2.26.0,<=2.31.0", optional = true }
-tqdm = { version = ">=4.62.3,<=4.66.2", optional = true }
+requests = { version = ">=2.26.0,<=2.32.3", optional = true }
+tqdm = { version = ">=4.62.3,<=4.66.4", optional = true }
 pillow = ">=9.4"

 [tool.poetry.extras]
@@ -51,10 +51,10 @@ desktop = ["opencv-python"]
 assets = ["requests","tqdm"]

 [tool.poetry.group.dev.dependencies]
-twine = ">=4.0.2,<6.0.0"
+twine = "^5.1.1"
 pytest = ">=7.2.2,<9.0.0"
 wheel = ">=0.40,<0.44"
-build = ">=0.10,<1.2"
+build = ">=0.10,<1.3"
 ruff = ">=0.1.0"
 mypy = "^1.4.1"
 pre-commit = "^3.3.3"
@@ -63,11 +63,15 @@ notebook = ">=6.5.3,<8.0.0"
 ipywidgets = "^8.1.1"
 jupytext = "^1.16.1"
 nbconvert = "^7.14.2"
+docutils = [
+    { version = "^0.20.1", python = "<3.9" },
+    { version = "^0.21.1", python = ">=3.9" }
+]


 [tool.poetry.group.docs.dependencies]
 mkdocs-material = {extras = ["imaging"], version = "^9.5.5"}
-mkdocstrings = {extras = ["python"], version = ">=0.20,<0.25"}
+mkdocstrings = {extras = ["python"], version = ">=0.20,<0.26"}
 mike = "^2.0.0"
 # For Documentation Development use Python 3.10 or above
 # Use Latest mkdocs-jupyter min 0.24.6 for Jupyter Notebook Theme support
@@ -23,6 +23,7 @@ from supervision.annotators.core import (
    PercentageBarAnnotator,
    PixelateAnnotator,
    PolygonAnnotator,
+    RichLabelAnnotator,
    RoundBoxAnnotator,
    TraceAnnotator,
    TriangleAnnotator,
@@ -34,9 +35,17 @@ from supervision.dataset.core import (
    ClassificationDataset,
    DetectionDataset,
 )
+from supervision.dataset.utils import mask_to_rle, rle_to_mask
 from supervision.detection.annotate import BoxAnnotator
 from supervision.detection.core import Detections
 from supervision.detection.line_zone import LineZone, LineZoneAnnotator
+from supervision.detection.lmm import LMM
+from supervision.detection.overlap_filter import (
+    OverlapFilter,
+    box_non_max_merge,
+    box_non_max_suppression,
+    mask_non_max_suppression,
+)
 from supervision.detection.tools.csv_sink import CSVSink
 from supervision.detection.tools.inference_slicer import InferenceSlicer
 from supervision.detection.tools.json_sink import JSONSink
@@ -44,22 +53,25 @@ from supervision.detection.tools.polygon_zone import PolygonZone, PolygonZoneAnn
 from supervision.detection.tools.smoother import DetectionsSmoother
 from supervision.detection.utils import (
    box_iou_batch,
-    box_non_max_suppression,
    calculate_masks_centroids,
+    clip_boxes,
+    contains_holes,
+    contains_multiple_segments,
    filter_polygons_by_area,
    mask_iou_batch,
-    mask_non_max_suppression,
    mask_to_polygons,
    mask_to_xyxy,
    move_boxes,
+    move_masks,
+    pad_boxes,
    polygon_to_mask,
    polygon_to_xyxy,
    scale_boxes,
 )
 from supervision.draw.color import Color, ColorPalette
 from supervision.draw.utils import (
-    calculate_dynamic_line_thickness,
-    calculate_dynamic_text_scale,
+    calculate_optimal_line_thickness,
+    calculate_optimal_text_scale,
    draw_filled_rectangle,
    draw_image,
    draw_line,
@@ -69,10 +81,25 @@ from supervision.draw.utils import (
 )
 from supervision.geometry.core import Point, Position, Rect
 from supervision.geometry.utils import get_polygon_center
+from supervision.keypoint.annotators import (
+    EdgeAnnotator,
+    VertexAnnotator,
+    VertexLabelAnnotator,
+)
+from supervision.keypoint.core import KeyPoints
 from supervision.metrics.detection import ConfusionMatrix, MeanAveragePrecision
 from supervision.tracker.byte_tracker.core import ByteTrack
+from supervision.utils.conversion import cv2_to_pillow, pillow_to_cv2
 from supervision.utils.file import list_files_with_extensions
-from supervision.utils.image import ImageSink, crop_image, place_image, resize_image
+from supervision.utils.image import (
+    ImageSink,
+    create_tiles,
+    crop_image,
+    letterbox_image,
+    overlay_image,
+    resize_image,
+    scale_image,
+)
 from supervision.utils.notebook import plot_image, plot_images_grid
 from supervision.utils.video import (
    FPSMonitor,
@@ -3,13 +3,14 @@ from typing import List, Optional, Tuple, Union

 import cv2
 import numpy as np
+from PIL import ImageDraw, ImageFont

 from supervision.annotators.base import BaseAnnotator, ImageType
 from supervision.annotators.utils import (
    ColorLookup,
    Trace,
    resolve_color,
-    scene_to_annotator_img_type,
+    resolve_text_background_xyxy,
 )
 from supervision.config import CLASS_NAME_DATA_FIELD, ORIENTED_BOX_COORDINATES
 from supervision.detection.core import Detections
@@ -17,7 +18,11 @@ from supervision.detection.utils import clip_boxes, mask_to_polygons
 from supervision.draw.color import Color, ColorPalette
 from supervision.draw.utils import draw_polygon
 from supervision.geometry.core import Position
-from supervision.utils.image import crop_image, place_image, resize_image
+from supervision.utils.conversion import (
+    ensure_cv2_image_for_annotation,
+    ensure_pil_image_for_annotation,
+)
+from supervision.utils.image import crop_image, overlay_image, scale_image


 class BoundingBoxAnnotator(BaseAnnotator):
@@ -43,7 +48,7 @@ class BoundingBoxAnnotator(BaseAnnotator):
        self.thickness: int = thickness
        self.color_lookup: ColorLookup = color_lookup

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -124,7 +129,7 @@ class OrientedBoxAnnotator(BaseAnnotator):
        self.thickness: int = thickness
        self.color_lookup: ColorLookup = color_lookup

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -170,7 +175,7 @@ class OrientedBoxAnnotator(BaseAnnotator):
            return scene

        for detection_idx in range(len(detections)):
-            bbox = np.int0(detections.data.get(ORIENTED_BOX_COORDINATES)[detection_idx])
+            bbox = np.intp(detections.data.get(ORIENTED_BOX_COORDINATES)[detection_idx])
            color = resolve_color(
                color=self.color,
                detections=detections,
@@ -212,7 +217,7 @@ class MaskAnnotator(BaseAnnotator):
        self.opacity = opacity
        self.color_lookup: ColorLookup = color_lookup

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -268,8 +273,10 @@ class MaskAnnotator(BaseAnnotator):
            mask = detections.mask[detection_idx]
            colored_mask[mask] = color.as_bgr()

-        scene = cv2.addWeighted(colored_mask, self.opacity, scene, 1 - self.opacity, 0)
-        return scene.astype(np.uint8)
+        cv2.addWeighted(
+            colored_mask, self.opacity, scene, 1 - self.opacity, 0, dst=scene
+        )
+        return scene


 class PolygonAnnotator(BaseAnnotator):
@@ -299,7 +306,7 @@ class PolygonAnnotator(BaseAnnotator):
        self.thickness: int = thickness
        self.color_lookup: ColorLookup = color_lookup

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -385,7 +392,7 @@ class ColorAnnotator(BaseAnnotator):
        self.color_lookup: ColorLookup = color_lookup
        self.opacity = opacity

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -424,7 +431,7 @@ class ColorAnnotator(BaseAnnotator):
        ![box-mask-annotator-example](https://media.roboflow.com/
        supervision-annotator-examples/box-mask-annotator-example-purple.png)
        """
-        mask_image = scene.copy()
+        scene_with_boxes = scene.copy()
        for detection_idx in range(len(detections)):
            x1, y1, x2, y2 = detections.xyxy[detection_idx].astype(int)
            color = resolve_color(
@@ -436,14 +443,15 @@ class ColorAnnotator(BaseAnnotator):
                else custom_color_lookup,
            )
            cv2.rectangle(
-                img=scene,
+                img=scene_with_boxes,
                pt1=(x1, y1),
                pt2=(x2, y2),
                color=color.as_bgr(),
                thickness=-1,
            )
-        scene = cv2.addWeighted(
-            scene, self.opacity, mask_image, 1 - self.opacity, gamma=0
+
+        cv2.addWeighted(
+            scene_with_boxes, self.opacity, scene, 1 - self.opacity, gamma=0, dst=scene
        )
        return scene

@@ -479,7 +487,7 @@ class HaloAnnotator(BaseAnnotator):
        self.color_lookup: ColorLookup = color_lookup
        self.kernel_size: int = kernel_size

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -544,7 +552,8 @@ class HaloAnnotator(BaseAnnotator):
        gray = cv2.cvtColor(colored_mask, cv2.COLOR_BGR2GRAY)
        alpha = self.opacity * gray / gray.max()
        alpha_mask = alpha[:, :, np.newaxis]
-        scene = np.uint8(scene * (1 - alpha_mask) + colored_mask * self.opacity)
+        blended_scene = np.uint8(scene * (1 - alpha_mask) + colored_mask * self.opacity)
+        np.copyto(scene, blended_scene)
        return scene


@@ -577,7 +586,7 @@ class EllipseAnnotator(BaseAnnotator):
        self.end_angle: int = end_angle
        self.color_lookup: ColorLookup = color_lookup

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -668,7 +677,7 @@ class BoxCornerAnnotator(BaseAnnotator):
        self.corner_length: int = corner_length
        self.color_lookup: ColorLookup = color_lookup

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -756,7 +765,7 @@ class CircleAnnotator(BaseAnnotator):
        self.thickness: int = thickness
        self.color_lookup: ColorLookup = color_lookup

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -831,6 +840,7 @@ class DotAnnotator(BaseAnnotator):
        radius: int = 4,
        position: Position = Position.CENTER,
        color_lookup: ColorLookup = ColorLookup.CLASS,
+        outline_thickness: int = 0,
    ):
        """
        Args:
@@ -840,13 +850,15 @@ class DotAnnotator(BaseAnnotator):
            position (Position): The anchor position for placing the dot.
            color_lookup (ColorLookup): Strategy for mapping colors to annotations.
                Options are `INDEX`, `CLASS`, `TRACK`.
+            outline_thickness (int): Thickness of the outline of the dot.
        """
        self.color: Union[Color, ColorPalette] = color
        self.radius: int = radius
        self.position: Position = position
        self.color_lookup: ColorLookup = color_lookup
+        self.outline_thickness = outline_thickness

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -896,11 +908,16 @@ class DotAnnotator(BaseAnnotator):
                else custom_color_lookup,
            )
            center = (int(xy[detection_idx, 0]), int(xy[detection_idx, 1]))
+
            cv2.circle(scene, center, self.radius, color.as_bgr(), -1)
+            if self.outline_thickness:
+                cv2.circle(
+                    scene, center, self.radius, (0, 0, 0), self.outline_thickness
+                )
        return scene


-class LabelAnnotator:
+class LabelAnnotator(BaseAnnotator):
    """
    A class for annotating labels on an image using provided detections.
    """
@@ -914,6 +931,7 @@ class LabelAnnotator:
        text_padding: int = 10,
        text_position: Position = Position.TOP_LEFT,
        color_lookup: ColorLookup = ColorLookup.CLASS,
+        border_radius: int = 0,
    ):
        """
        Args:
@@ -927,7 +945,10 @@ class LabelAnnotator:
                Possible values are defined in the `Position` enum.
            color_lookup (str): Strategy for mapping colors to annotations.
                Options are `INDEX`, `CLASS`, `TRACK`.
+            border_radius (int): The radius to apply round edges. If the selected
+                value is higher than the lower dimension, width or height, is clipped.
        """
+        self.border_radius: int = border_radius
        self.color: Union[Color, ColorPalette] = color
        self.text_color: Color = text_color
        self.text_scale: float = text_scale
@@ -936,65 +957,12 @@ class LabelAnnotator:
        self.text_anchor: Position = text_position
        self.color_lookup: ColorLookup = color_lookup

-    @staticmethod
-    def resolve_text_background_xyxy(
-        center_coordinates: Tuple[int, int],
-        text_wh: Tuple[int, int],
-        position: Position,
-    ) -> Tuple[int, int, int, int]:
-        center_x, center_y = center_coordinates
-        text_w, text_h = text_wh
-
-        if position == Position.TOP_LEFT:
-            return center_x, center_y - text_h, center_x + text_w, center_y
-        elif position == Position.TOP_RIGHT:
-            return center_x - text_w, center_y - text_h, center_x, center_y
-        elif position == Position.TOP_CENTER:
-            return (
-                center_x - text_w // 2,
-                center_y - text_h,
-                center_x + text_w // 2,
-                center_y,
-            )
-        elif position == Position.CENTER or position == Position.CENTER_OF_MASS:
-            return (
-                center_x - text_w // 2,
-                center_y - text_h // 2,
-                center_x + text_w // 2,
-                center_y + text_h // 2,
-            )
-        elif position == Position.BOTTOM_LEFT:
-            return center_x, center_y, center_x + text_w, center_y + text_h
-        elif position == Position.BOTTOM_RIGHT:
-            return center_x - text_w, center_y, center_x, center_y + text_h
-        elif position == Position.BOTTOM_CENTER:
-            return (
-                center_x - text_w // 2,
-                center_y,
-                center_x + text_w // 2,
-                center_y + text_h,
-            )
-        elif position == Position.CENTER_LEFT:
-            return (
-                center_x - text_w,
-                center_y - text_h // 2,
-                center_x,
-                center_y + text_h // 2,
-            )
-        elif position == Position.CENTER_RIGHT:
-            return (
-                center_x,
-                center_y - text_h // 2,
-                center_x + text_w,
-                center_y + text_h // 2,
-            )
-
-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
        detections: Detections,
-        labels: List[str] = None,
+        labels: Optional[List[str]] = None,
        custom_color_lookup: Optional[np.ndarray] = None,
    ) -> ImageType:
        """
@@ -1005,7 +973,7 @@ class LabelAnnotator:
                `ImageType` is a flexible type, accepting either `numpy.ndarray`
                or `PIL.Image.Image`.
            detections (Detections): Object detections to annotate.
-            labels (List[str]): Optional. Custom labels for each detection.
+            labels (Optional[List[str]]): Custom labels for each detection.
            custom_color_lookup (Optional[np.ndarray]): Custom color lookup array.
                Allows to override the default color mapping strategy.

@@ -1020,10 +988,17 @@ class LabelAnnotator:
            image = ...
            detections = sv.Detections(...)

+            labels = [
+                f"{class_name} {confidence:.2f}"
+                for class_name, confidence
+                in zip(detections['class_name'], detections.confidence)
+            ]
+
            label_annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER)
            annotated_frame = label_annotator.annotate(
                scene=image.copy(),
-                detections=detections
+                detections=detections,
+                labels=labels
            )
            ```

@@ -1036,12 +1011,9 @@ class LabelAnnotator:
        ).astype(int)
        if labels is not None and len(labels) != len(detections):
            raise ValueError(
-                f"The number of labels provided ({len(labels)}) does not match the "
-                f"number of detections ({len(detections)}). Each detection should have "
-                f"a corresponding label. This discrepancy can occur if the labels and "
-                f"detections are not aligned or if an incorrect number of labels has "
-                f"been provided. Please ensure that the labels array has the same "
-                f"length as the Detections object."
+                f"The number of labels ({len(labels)}) does not match the "
+                f"number of detections ({len(detections)}). Each detection "
+                f"should have exactly 1 label."
            )

        for detection_idx, center_coordinates in enumerate(anchors_coordinates):
@@ -1049,9 +1021,11 @@ class LabelAnnotator:
                color=self.color,
                detections=detections,
                detection_idx=detection_idx,
-                color_lookup=self.color_lookup
-                if custom_color_lookup is None
-                else custom_color_lookup,
+                color_lookup=(
+                    self.color_lookup
+                    if custom_color_lookup is None
+                    else custom_color_lookup
+                ),
            )

            if labels is not None:
@@ -1071,7 +1045,7 @@ class LabelAnnotator:
            )[0]
            text_w_padded = text_w + 2 * self.text_padding
            text_h_padded = text_h + 2 * self.text_padding
-            text_background_xyxy = self.resolve_text_background_xyxy(
+            text_background_xyxy = resolve_text_background_xyxy(
                center_coordinates=tuple(center_coordinates),
                text_wh=(text_w_padded, text_h_padded),
                position=self.text_anchor,
@@ -1080,12 +1054,11 @@ class LabelAnnotator:
            text_x = text_background_xyxy[0] + self.text_padding
            text_y = text_background_xyxy[1] + self.text_padding + text_h

-            cv2.rectangle(
-                img=scene,
-                pt1=(text_background_xyxy[0], text_background_xyxy[1]),
-                pt2=(text_background_xyxy[2], text_background_xyxy[3]),
+            self.draw_rounded_rectangle(
+                scene=scene,
+                xyxy=text_background_xyxy,
                color=color.as_bgr(),
-                thickness=cv2.FILLED,
+                border_radius=self.border_radius,
            )
            cv2.putText(
                img=scene,
@@ -1099,6 +1072,215 @@ class LabelAnnotator:
            )
        return scene

+    @staticmethod
+    def draw_rounded_rectangle(
+        scene: np.ndarray,
+        xyxy: Tuple[int, int, int, int],
+        color: Tuple[int, int, int],
+        border_radius: int,
+    ) -> np.ndarray:
+        x1, y1, x2, y2 = xyxy
+        width = x2 - x1
+        height = y2 - y1
+
+        border_radius = min(border_radius, min(width, height) // 2)
+
+        rectangle_coordinates = [
+            ((x1 + border_radius, y1), (x2 - border_radius, y2)),
+            ((x1, y1 + border_radius), (x2, y2 - border_radius)),
+        ]
+        circle_centers = [
+            (x1 + border_radius, y1 + border_radius),
+            (x2 - border_radius, y1 + border_radius),
+            (x1 + border_radius, y2 - border_radius),
+            (x2 - border_radius, y2 - border_radius),
+        ]
+
+        for coordinates in rectangle_coordinates:
+            cv2.rectangle(
+                img=scene,
+                pt1=coordinates[0],
+                pt2=coordinates[1],
+                color=color,
+                thickness=-1,
+            )
+        for center in circle_centers:
+            cv2.circle(
+                img=scene,
+                center=center,
+                radius=border_radius,
+                color=color,
+                thickness=-1,
+            )
+        return scene
+
+
+class RichLabelAnnotator(BaseAnnotator):
+    """
+    A class for annotating labels on an image using provided detections,
+    with support for Unicode characters by using a custom font.
+    """
+
+    def __init__(
+        self,
+        color: Union[Color, ColorPalette] = ColorPalette.DEFAULT,
+        text_color: Color = Color.WHITE,
+        font_path: Optional[str] = None,
+        font_size: int = 10,
+        text_padding: int = 10,
+        text_position: Position = Position.TOP_LEFT,
+        color_lookup: ColorLookup = ColorLookup.CLASS,
+        border_radius: int = 0,
+    ):
+        """
+        Args:
+            color (Union[Color, ColorPalette]): The color or color palette to use for
+                annotating the text background.
+            text_color (Color): The color to use for the text.
+            font_path (Optional[str]): Path to the font file (e.g., ".ttf" or ".otf")
+                to use for rendering text. If `None`, the default PIL font will be used.
+            font_size (int): Font size for the text.
+            text_padding (int): Padding around the text within its background box.
+            text_position (Position): Position of the text relative to the detection.
+                Possible values are defined in the `Position` enum.
+            color_lookup (ColorLookup): Strategy for mapping colors to annotations.
+                Options are `INDEX`, `CLASS`, `TRACK`.
+            border_radius (int): The radius to apply round edges. If the selected
+                value is higher than the lower dimension, width or height, is clipped.
+        """
+        self.color = color
+        self.text_color = text_color
+        self.text_padding = text_padding
+        self.text_anchor = text_position
+        self.color_lookup = color_lookup
+        self.border_radius = border_radius
+        if font_path is not None:
+            try:
+                self.font = ImageFont.truetype(font_path, font_size)
+            except OSError:
+                print(f"Font path '{font_path}' not found. Using PIL's default font.")
+                self.font = self._load_default_font(font_size)
+        else:
+            self.font = self._load_default_font(font_size)
+
+    @ensure_pil_image_for_annotation
+    def annotate(
+        self,
+        scene: ImageType,
+        detections: Detections,
+        labels: Optional[List[str]] = None,
+        custom_color_lookup: Optional[np.ndarray] = None,
+    ) -> ImageType:
+        """
+        Annotates the given scene with labels based on the provided
+        detections, with support for Unicode characters.
+
+        Args:
+            scene (ImageType): The image where labels will be drawn.
+                `ImageType` is a flexible type, accepting either `numpy.ndarray`
+                or `PIL.Image.Image`.
+            detections (Detections): Object detections to annotate.
+            labels (Optional[List[str]]): Custom labels for each detection.
+            custom_color_lookup (Optional[np.ndarray]): Custom color lookup array.
+                Allows to override the default color mapping strategy.
+
+        Returns:
+            The annotated image, matching the type of `scene` (`numpy.ndarray`
+                or `PIL.Image.Image`)
+
+        Example:
+            ```python
+            import supervision as sv
+
+            image = ...
+            detections = sv.Detections(...)
+
+            labels = [
+                f"{class_name} {confidence:.2f}"
+                for class_name, confidence
+                in zip(detections['class_name'], detections.confidence)
+            ]
+
+            rich_label_annotator = sv.RichLabelAnnotator(font_path="path/to/font.ttf")
+            annotated_frame = label_annotator.annotate(
+                scene=image.copy(),
+                detections=detections,
+                labels=labels
+            )
+            ```
+
+        """
+        draw = ImageDraw.Draw(scene)
+        anchors_coordinates = detections.get_anchors_coordinates(
+            anchor=self.text_anchor
+        ).astype(int)
+        if labels is not None and len(labels) != len(detections):
+            raise ValueError(
+                f"The number of labels provided ({len(labels)}) does not match the "
+                f"number of detections ({len(detections)}). Each detection should have "
+                f"a corresponding label."
+            )
+        for detection_idx, center_coordinates in enumerate(anchors_coordinates):
+            color = resolve_color(
+                color=self.color,
+                detections=detections,
+                detection_idx=detection_idx,
+                color_lookup=(
+                    self.color_lookup
+                    if custom_color_lookup is None
+                    else custom_color_lookup
+                ),
+            )
+            if labels is not None:
+                text = labels[detection_idx]
+            elif detections[CLASS_NAME_DATA_FIELD] is not None:
+                text = detections[CLASS_NAME_DATA_FIELD][detection_idx]
+            elif detections.class_id is not None:
+                text = str(detections.class_id[detection_idx])
+            else:
+                text = str(detection_idx)
+
+            left, top, right, bottom = draw.textbbox((0, 0), text, font=self.font)
+            text_width = right - left
+            text_height = bottom - top
+            text_w_padded = text_width + 2 * self.text_padding
+            text_h_padded = text_height + 2 * self.text_padding
+            text_background_xyxy = resolve_text_background_xyxy(
+                center_coordinates=tuple(center_coordinates),
+                text_wh=(text_w_padded, text_h_padded),
+                position=self.text_anchor,
+            )
+
+            text_x = text_background_xyxy[0] + self.text_padding - left
+            text_y = text_background_xyxy[1] + self.text_padding - top
+
+            draw.rounded_rectangle(
+                text_background_xyxy,
+                radius=self.border_radius,
+                fill=color.as_rgb(),
+                outline=None,
+            )
+            draw.text(
+                xy=(text_x, text_y),
+                text=text,
+                font=self.font,
+                fill=self.text_color.as_rgb(),
+            )
+        return scene
+
+    @staticmethod
+    def _load_default_font(size):
+        """
+        PIL either loads a font that accepts a size (e.g. on my machine)
+        or raises an error saying `load_default` does not accept arguments
+        (e.g. in Colab).
+        """
+        try:
+            font = ImageFont.load_default(size)
+        except TypeError:
+            font = ImageFont.load_default()
+        return font
+

 class BlurAnnotator(BaseAnnotator):
    """
@@ -1112,7 +1294,7 @@ class BlurAnnotator(BaseAnnotator):
        """
        self.kernel_size: int = kernel_size

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -1161,7 +1343,7 @@ class BlurAnnotator(BaseAnnotator):
        return scene


-class TraceAnnotator:
+class TraceAnnotator(BaseAnnotator):
    """
    A class for drawing trace paths on an image based on detection coordinates.

@@ -1197,7 +1379,7 @@ class TraceAnnotator:
        self.thickness = thickness
        self.color_lookup: ColorLookup = color_lookup

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -1270,7 +1452,7 @@ class TraceAnnotator:
        return scene


-class HeatMapAnnotator:
+class HeatMapAnnotator(BaseAnnotator):
    """
    A class for drawing heatmaps on an image based on provided detections.
    Heat accumulates over time and is drawn as a semi-transparent overlay
@@ -1304,7 +1486,7 @@ class HeatMapAnnotator:
        self.top_hue = top_hue
        self.low_hue = low_hue

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(self, scene: ImageType, detections: Detections) -> ImageType:
        """
        Annotates the scene with a heatmap based on the provided detections.
@@ -1380,7 +1562,7 @@ class PixelateAnnotator(BaseAnnotator):
        """
        self.pixel_size: int = pixel_size

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -1451,6 +1633,7 @@ class TriangleAnnotator(BaseAnnotator):
        height: int = 10,
        position: Position = Position.TOP_CENTER,
        color_lookup: ColorLookup = ColorLookup.CLASS,
+        outline_thickness: int = 0,
    ):
        """
        Args:
@@ -1461,14 +1644,16 @@ class TriangleAnnotator(BaseAnnotator):
            position (Position): The anchor position for placing the triangle.
            color_lookup (ColorLookup): Strategy for mapping colors to annotations.
                Options are `INDEX`, `CLASS`, `TRACK`.
+            outline_thickness (int): Thickness of the outline of the triangle.
        """
        self.color: Union[Color, ColorPalette] = color
        self.base: int = base
        self.height: int = height
        self.position: Position = position
        self.color_lookup: ColorLookup = color_lookup
+        self.outline_thickness: int = outline_thickness

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -1528,7 +1713,10 @@ class TriangleAnnotator(BaseAnnotator):
            )

            cv2.fillPoly(scene, [vertices], color.as_bgr())
-
+            if self.outline_thickness:
+                cv2.polylines(
+                    scene, [vertices], True, (0, 0, 0), thickness=self.outline_thickness
+                )
        return scene


@@ -1564,7 +1752,7 @@ class RoundBoxAnnotator(BaseAnnotator):
            raise ValueError("roundness attribute must be float between (0, 1.0]")
        self.roundness: float = roundness

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -1701,7 +1889,7 @@ class PercentageBarAnnotator(BaseAnnotator):
        if border_thickness is None:
            self.border_thickness = int(0.15 * self.height)

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -1847,7 +2035,14 @@ class CropAnnotator(BaseAnnotator):
    A class for drawing scaled up crops of detections on the scene.
    """

-    def __init__(self, position: Position = Position.TOP_CENTER, scale_factor: int = 2):
+    def __init__(
+        self,
+        position: Position = Position.TOP_CENTER,
+        scale_factor: int = 2,
+        border_color: Union[Color, ColorPalette] = ColorPalette.DEFAULT,
+        border_thickness: int = 2,
+        border_color_lookup: ColorLookup = ColorLookup.CLASS,
+    ):
        """
        Args:
            position (Position): The anchor position for placing the cropped and scaled
@@ -1855,16 +2050,25 @@ class CropAnnotator(BaseAnnotator):
            scale_factor (int): The factor by which to scale the cropped image part. A
                factor of 2, for example, would double the size of the cropped area,
                allowing for a closer view of the detection.
+            border_color (Union[Color, ColorPalette]): The color or color palette to
+                use for annotating border around the cropped area.
+            border_thickness (int): The thickness of the border around the cropped area.
+            border_color_lookup (ColorLookup): Strategy for mapping colors to
+                annotations. Options are `INDEX`, `CLASS`, `TRACK`.
        """
        self.position: Position = position
        self.scale_factor: int = scale_factor
+        self.border_color: Union[Color, ColorPalette] = border_color
+        self.border_thickness: int = border_thickness
+        self.border_color_lookup: ColorLookup = border_color_lookup

-    @scene_to_annotator_img_type
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
-        scene: np.ndarray,
+        scene: ImageType,
        detections: Detections,
-    ) -> np.ndarray:
+        custom_color_lookup: Optional[np.ndarray] = None,
+    ) -> ImageType:
        """
        Annotates the provided scene with scaled and cropped parts of the image based
        on the provided detections. Each detection is cropped from the original scene
@@ -1873,8 +2077,12 @@ class CropAnnotator(BaseAnnotator):


        Args:
-            scene (np.ndarray): The image where cropped detection will be placed.
+            scene (ImageType): The image where cropped detection will be placed.
+                `ImageType` is a flexible type, accepting either `numpy.ndarray`
+                or `PIL.Image.Image`.
            detections (Detections): Object detections to annotate.
+            custom_color_lookup (Optional[np.ndarray]): Custom color lookup array.
+                Allows to override the default color mapping strategy.

        Returns:
            The annotated image.
@@ -1897,41 +2105,71 @@ class CropAnnotator(BaseAnnotator):
            crop_image(image=scene, xyxy=xyxy) for xyxy in detections.xyxy.astype(int)
        ]
        resized_crops = [
-            resize_image(image=crop, scale_factor=self.scale_factor) for crop in crops
+            scale_image(image=crop, scale_factor=self.scale_factor) for crop in crops
        ]
        anchors = detections.get_anchors_coordinates(anchor=self.position).astype(int)

-        for resized_crop, anchor in zip(resized_crops, anchors):
+        for idx, (resized_crop, anchor) in enumerate(zip(resized_crops, anchors)):
            crop_wh = resized_crop.shape[1], resized_crop.shape[0]
-            crop_anchor = self.calculate_crop_coordinates(
+            (x1, y1), (x2, y2) = self.calculate_crop_coordinates(
                anchor=anchor, crop_wh=crop_wh, position=self.position
            )
-            scene = place_image(scene=scene, image=resized_crop, anchor=crop_anchor)
+            scene = overlay_image(image=scene, overlay=resized_crop, anchor=(x1, y1))
+            color = resolve_color(
+                color=self.border_color,
+                detections=detections,
+                detection_idx=idx,
+                color_lookup=self.border_color_lookup
+                if custom_color_lookup is None
+                else custom_color_lookup,
+            )
+            cv2.rectangle(
+                img=scene,
+                pt1=(x1, y1),
+                pt2=(x2, y2),
+                color=color.as_bgr(),
+                thickness=self.border_thickness,
+            )

        return scene

    @staticmethod
    def calculate_crop_coordinates(
        anchor: Tuple[int, int], crop_wh: Tuple[int, int], position: Position
-    ) -> Tuple[int, int]:
+    ) -> Tuple[Tuple[int, int], Tuple[int, int]]:
        anchor_x, anchor_y = anchor
        width, height = crop_wh

        if position == Position.TOP_LEFT:
-            return anchor_x - width, anchor_y - height
+            return (anchor_x - width, anchor_y - height), (anchor_x, anchor_y)
        elif position == Position.TOP_CENTER:
-            return anchor_x - width // 2, anchor_y - height
+            return (
+                (anchor_x - width // 2, anchor_y - height),
+                (anchor_x + width // 2, anchor_y),
+            )
        elif position == Position.TOP_RIGHT:
-            return anchor_x, anchor_y - height
+            return (anchor_x, anchor_y - height), (anchor_x + width, anchor_y)
        elif position == Position.CENTER_LEFT:
-            return anchor_x - width, anchor_y - height // 2
+            return (
+                (anchor_x - width, anchor_y - height // 2),
+                (anchor_x, anchor_y + height // 2),
+            )
        elif position == Position.CENTER or position == Position.CENTER_OF_MASS:
-            return anchor_x - width // 2, anchor_y - height // 2
+            return (
+                (anchor_x - width // 2, anchor_y - height // 2),
+                (anchor_x + width // 2, anchor_y + height // 2),
+            )
        elif position == Position.CENTER_RIGHT:
-            return anchor_x, anchor_y - height // 2
+            return (
+                (anchor_x, anchor_y - height // 2),
+                (anchor_x + width, anchor_y + height // 2),
+            )
        elif position == Position.BOTTOM_LEFT:
-            return anchor_x - width, anchor_y
+            return (anchor_x - width, anchor_y), (anchor_x, anchor_y + height)
        elif position == Position.BOTTOM_CENTER:
-            return anchor_x - width // 2, anchor_y
+            return (
+                (anchor_x - width // 2, anchor_y),
+                (anchor_x + width // 2, anchor_y + height),
+            )
        elif position == Position.BOTTOM_RIGHT:
-            return anchor_x, anchor_y
+            return (anchor_x, anchor_y), (anchor_x + width, anchor_y + height)
@@ -1,12 +1,8 @@
 from enum import Enum
-from functools import wraps
-from typing import Optional, Union
+from typing import Optional, Tuple, Union

-import cv2
 import numpy as np
-from PIL import Image

-from supervision.annotators.base import ImageType
 from supervision.detection.core import Detections
 from supervision.draw.color import Color, ColorPalette
 from supervision.geometry.core import Position
@@ -38,14 +34,14 @@ def resolve_color_idx(
 ) -> int:
    if detection_idx >= len(detections):
        raise ValueError(
-            f"Detection index {detection_idx}"
+            f"Detection index {detection_idx} "
            f"is out of bounds for detections of length {len(detections)}"
        )

    if isinstance(color_lookup, np.ndarray):
        if len(color_lookup) != len(detections):
            raise ValueError(
-                f"Length of color lookup {len(color_lookup)}"
+                f"Length of color lookup {len(color_lookup)} "
                f"does not match length of detections {len(detections)}"
            )
        return color_lookup[detection_idx]
@@ -54,19 +50,72 @@ def resolve_color_idx(
    elif color_lookup == ColorLookup.CLASS:
        if detections.class_id is None:
            raise ValueError(
-                "Could not resolve color by class because"
+                "Could not resolve color by class because "
                "Detections do not have class_id"
            )
        return detections.class_id[detection_idx]
    elif color_lookup == ColorLookup.TRACK:
        if detections.tracker_id is None:
            raise ValueError(
-                "Could not resolve color by track because"
+                "Could not resolve color by track because "
                "Detections do not have tracker_id"
            )
        return detections.tracker_id[detection_idx]


+def resolve_text_background_xyxy(
+    center_coordinates: Tuple[int, int],
+    text_wh: Tuple[int, int],
+    position: Position,
+) -> Tuple[int, int, int, int]:
+    center_x, center_y = center_coordinates
+    text_w, text_h = text_wh
+
+    if position == Position.TOP_LEFT:
+        return center_x, center_y - text_h, center_x + text_w, center_y
+    elif position == Position.TOP_RIGHT:
+        return center_x - text_w, center_y - text_h, center_x, center_y
+    elif position == Position.TOP_CENTER:
+        return (
+            center_x - text_w // 2,
+            center_y - text_h,
+            center_x + text_w // 2,
+            center_y,
+        )
+    elif position == Position.CENTER or position == Position.CENTER_OF_MASS:
+        return (
+            center_x - text_w // 2,
+            center_y - text_h // 2,
+            center_x + text_w // 2,
+            center_y + text_h // 2,
+        )
+    elif position == Position.BOTTOM_LEFT:
+        return center_x, center_y, center_x + text_w, center_y + text_h
+    elif position == Position.BOTTOM_RIGHT:
+        return center_x - text_w, center_y, center_x, center_y + text_h
+    elif position == Position.BOTTOM_CENTER:
+        return (
+            center_x - text_w // 2,
+            center_y,
+            center_x + text_w // 2,
+            center_y + text_h,
+        )
+    elif position == Position.CENTER_LEFT:
+        return (
+            center_x - text_w,
+            center_y - text_h // 2,
+            center_x,
+            center_y + text_h // 2,
+        )
+    elif position == Position.CENTER_RIGHT:
+        return (
+            center_x,
+            center_y - text_h // 2,
+            center_x + text_w,
+            center_y + text_h // 2,
+        )
+
+
 def get_color_by_index(color: Union[Color, ColorPalette], idx: int) -> Color:
    if isinstance(color, ColorPalette):
        return color.by_idx(idx)
@@ -123,33 +172,3 @@ class Trace:

    def get(self, tracker_id: int) -> np.ndarray:
        return self.xy[self.tracker_id == tracker_id]
-
-
-def pillow_to_cv2(image: Image.Image) -> np.ndarray:
-    scene = np.array(image)
-    scene = cv2.cvtColor(scene, cv2.COLOR_RGB2BGR)
-    return scene
-
-
-def scene_to_annotator_img_type(annotate_func):
-    """
-    Decorates `BaseAnnotator.annotate` implementations, converts scene to
-    an image type used internally by the annotators, converts back when annotation
-    is complete.
-    """
-
-    @wraps(annotate_func)
-    def wrapper(self, scene: ImageType, *args, **kwargs):
-        if isinstance(scene, np.ndarray):
-            return annotate_func(self, scene, *args, **kwargs)
-
-        if isinstance(scene, Image.Image):
-            scene = pillow_to_cv2(scene)
-            annotated = annotate_func(self, scene, *args, **kwargs)
-            annotated = cv2.cvtColor(annotated, cv2.COLOR_BGR2RGB)
-            annotated = Image.fromarray(annotated)
-            return annotated
-
-        raise ValueError(f"Unsupported image type: {type(scene)}")
-
-    return wrapper
@@ -18,6 +18,8 @@ class VideoAssets(Enum):
    | `SUBWAY`               | `subway.mp4`               | [Link](https://media.roboflow.com/supervision/video-examples/subway.mp4)              |
    | `MARKET_SQUARE`        | `market-square.mp4`        | [Link](https://media.roboflow.com/supervision/video-examples/market-square.mp4)       |
    | `PEOPLE_WALKING`       | `people-walking.mp4`       | [Link](https://media.roboflow.com/supervision/video-examples/people-walking.mp4)      |
+    | `BEACH`                | `beach-1.mp4`              | [Link](https://media.roboflow.com/supervision/video-examples/beach-1.mp4)             |
+    | `BASKETBALL`           | `basketball-1.mp4`         | [Link](https://media.roboflow.com/supervision/video-examples/basketball-1.mp4)        |
    """  # noqa: E501 // docs

    VEHICLES = "vehicles.mp4"
@@ -27,6 +29,8 @@ class VideoAssets(Enum):
    SUBWAY = "subway.mp4"
    MARKET_SQUARE = "market-square.mp4"
    PEOPLE_WALKING = "people-walking.mp4"
+    BEACH = "beach-1.mp4"
+    BASKETBALL = "basketball-1.mp4"

    @classmethod
    def list(cls):
@@ -62,4 +66,12 @@ VIDEO_ASSETS: Dict[str, Tuple[str, str]] = {
        f"{BASE_VIDEO_URL}{VideoAssets.PEOPLE_WALKING.value}",
        "0574c053c8686c3f1dc0aa3743e45cb9",
    ),
+    VideoAssets.BEACH.value: (
+        f"{BASE_VIDEO_URL}{VideoAssets.BEACH.value}",
+        "4175d42fec4d450ed081523fd39e0cf8",
+    ),
+    VideoAssets.BASKETBALL.value: (
+        f"{BASE_VIDEO_URL}{VideoAssets.BASKETBALL.value}",
+        "60d94a3c7c47d16f09d342b088012ecc",
+    ),
 }
@@ -116,13 +116,12 @@ class DetectionDataset(BaseDataset):
            Tuple[DetectionDataset, DetectionDataset]: A tuple containing
                the training and testing datasets.

-        Example:
+        Examples:
            ```python
            import supervision as sv

            ds = sv.DetectionDataset(...)
-            train_ds, test_ds = ds.split(split_ratio=0.7,
-                                         random_state=42, shuffle=True)
+            train_ds, test_ds = ds.split(split_ratio=0.7, random_state=42, shuffle=True)
            len(train_ds), len(test_ds)
            # (700, 300)
            ```
@@ -229,7 +228,7 @@ class DetectionDataset(BaseDataset):
            DetectionDataset: A DetectionDataset instance containing
                the loaded images and annotations.

-        Example:
+        Examples:
            ```python
            import roboflow
            from roboflow import Roboflow
@@ -267,6 +266,7 @@ class DetectionDataset(BaseDataset):
        annotations_directory_path: str,
        data_yaml_path: str,
        force_masks: bool = False,
+        is_obb: bool = False,
    ) -> DetectionDataset:
        """
        Creates a Dataset instance from YOLO formatted data.
@@ -281,12 +281,15 @@ class DetectionDataset(BaseDataset):
            force_masks (bool, optional): If True, forces
                masks to be loaded for all annotations,
                regardless of whether they are present.
+            is_obb (bool, optional): If True, loads the annotations in OBB format.
+                OBB annotations are defined as `[class_id, x, y, x, y, x, y, x, y]`,
+                where pairs of [x, y] are box corners.

        Returns:
            DetectionDataset: A DetectionDataset instance
                containing the loaded images and annotations.

-        Example:
+        Examples:
            ```python
            import roboflow
            from roboflow import Roboflow
@@ -313,6 +316,7 @@ class DetectionDataset(BaseDataset):
            annotations_directory_path=annotations_directory_path,
            data_yaml_path=data_yaml_path,
            force_masks=force_masks,
+            is_obb=is_obb,
        )
        return DetectionDataset(classes=classes, images=images, annotations=annotations)

@@ -391,7 +395,7 @@ class DetectionDataset(BaseDataset):
            DetectionDataset: A DetectionDataset instance containing
                the loaded images and annotations.

-        Example:
+        Examples:
            ```python
            import roboflow
            from roboflow import Roboflow
@@ -431,6 +435,20 @@ class DetectionDataset(BaseDataset):
        Exports the dataset to COCO format. This method saves the
        images and their corresponding annotations in COCO format.

+        !!! tip
+
+            The format of the mask is determined automatically based on its structure:
+
+            - If a mask contains multiple disconnected components or holes, it will be
+            saved using the Run-Length Encoding (RLE) format for efficient storage and
+            processing.
+            - If a mask consists of a single, contiguous region without any holes, it
+            will be encoded as a polygon, preserving the outline of the object.
+
+            This automatic selection ensures that the masks are stored in the most
+            appropriate and space-efficient format, complying with COCO dataset
+            standards.
+
        Args:
            images_directory_path (Optional[str]): The path to the directory
                where the images should be saved.
@@ -482,7 +500,7 @@ class DetectionDataset(BaseDataset):
            (DetectionDataset): A single `DetectionDataset` object containing
            the merged data from the input list.

-        Example:
+        Examples:
            ```python
            import supervision as sv

@@ -567,13 +585,12 @@ class ClassificationDataset(BaseDataset):
            Tuple[ClassificationDataset, ClassificationDataset]: A tuple containing
            the training and testing datasets.

-        Example:
+        Examples:
            ```python
            import supervision as sv

            cd = sv.ClassificationDataset(...)
-            train_cd,test_cd = cd.split(split_ratio=0.7,
-                                        random_state=42,shuffle=True)
+            train_cd,test_cd = cd.split(split_ratio=0.7, random_state=42,shuffle=True)
            len(train_cd), len(test_cd)
            # (700, 300)
            ```
@@ -635,7 +652,7 @@ class ClassificationDataset(BaseDataset):
        Returns:
            ClassificationDataset: The dataset.

-        Example:
+        Examples:
            ```python
            import roboflow
            from roboflow import Roboflow
@@ -5,13 +5,20 @@ from typing import Dict, List, Tuple

 import cv2
 import numpy as np
+import numpy.typing as npt

 from supervision.dataset.utils import (
    approximate_mask_with_polygons,
    map_detections_class_id,
+    mask_to_rle,
+    rle_to_mask,
 )
 from supervision.detection.core import Detections
-from supervision.detection.utils import polygon_to_mask
+from supervision.detection.utils import (
+    contains_holes,
+    contains_multiple_segments,
+    polygon_to_mask,
+)
 from supervision.utils.file import read_json_file, save_json_file


@@ -57,13 +64,24 @@ def group_coco_annotations_by_image_id(
    return annotations


-def _polygons_to_masks(
-    polygons: List[np.ndarray], resolution_wh: Tuple[int, int]
-) -> np.ndarray:
+def coco_annotations_to_masks(
+    image_annotations: List[dict], resolution_wh: Tuple[int, int]
+) -> npt.NDArray[np.bool_]:
    return np.array(
        [
-            polygon_to_mask(polygon=polygon, resolution_wh=resolution_wh)
-            for polygon in polygons
+            rle_to_mask(
+                rle=np.array(image_annotation["segmentation"]["counts"]),
+                resolution_wh=resolution_wh,
+            )
+            if image_annotation["iscrowd"]
+            else polygon_to_mask(
+                polygon=np.reshape(
+                    np.asarray(image_annotation["segmentation"], dtype=np.int32),
+                    (-1, 2),
+                ),
+                resolution_wh=resolution_wh,
+            )
+            for image_annotation in image_annotations
        ],
        dtype=bool,
    )
@@ -83,13 +101,9 @@ def coco_annotations_to_detections(
    xyxy[:, 2:4] += xyxy[:, 0:2]

    if with_masks:
-        polygons = [
-            np.reshape(
-                np.asarray(image_annotation["segmentation"], dtype=np.int32), (-1, 2)
-            )
-            for image_annotation in image_annotations
-        ]
-        mask = _polygons_to_masks(polygons=polygons, resolution_wh=resolution_wh)
+        mask = coco_annotations_to_masks(
+            image_annotations=image_annotations, resolution_wh=resolution_wh
+        )
        return Detections(
            class_id=np.asarray(class_ids, dtype=int), xyxy=xyxy, mask=mask
        )
@@ -108,24 +122,35 @@ def detections_to_coco_annotations(
    coco_annotations = []
    for xyxy, mask, _, class_id, _, _ in detections:
        box_width, box_height = xyxy[2] - xyxy[0], xyxy[3] - xyxy[1]
-        polygon = []
+        segmentation = []
+        iscrowd = 0
        if mask is not None:
-            polygon = list(
-                approximate_mask_with_polygons(
-                    mask=mask,
-                    min_image_area_percentage=min_image_area_percentage,
-                    max_image_area_percentage=max_image_area_percentage,
-                    approximation_percentage=approximation_percentage,
-                )[0].flatten()
-            )
+            iscrowd = contains_holes(mask=mask) or contains_multiple_segments(mask=mask)
+
+            if iscrowd:
+                segmentation = {
+                    "counts": mask_to_rle(mask=mask),
+                    "size": list(mask.shape[:2]),
+                }
+            else:
+                segmentation = [
+                    list(
+                        approximate_mask_with_polygons(
+                            mask=mask,
+                            min_image_area_percentage=min_image_area_percentage,
+                            max_image_area_percentage=max_image_area_percentage,
+                            approximation_percentage=approximation_percentage,
+                        )[0].flatten()
+                    )
+                ]
        coco_annotation = {
            "id": annotation_id,
            "image_id": image_id,
            "category_id": int(class_id),
            "bbox": [xyxy[0], xyxy[1], box_width, box_height],
            "area": box_width * box_height,
-            "segmentation": [polygon] if polygon else [],
-            "iscrowd": 0,
+            "segmentation": segmentation,
+            "iscrowd": iscrowd,
        }
        coco_annotations.append(coco_annotation)
        annotation_id += 1
@@ -5,6 +5,7 @@ from typing import Dict, List, Optional, Tuple
 import cv2
 import numpy as np

+from supervision.config import ORIENTED_BOX_COORDINATES
 from supervision.dataset.utils import approximate_mask_with_polygons
 from supervision.detection.core import Detections
 from supervision.detection.utils import polygon_to_mask, polygon_to_xyxy
@@ -70,12 +71,15 @@ def _image_name_to_annotation_name(image_name: str) -> str:


 def yolo_annotations_to_detections(
-    lines: List[str], resolution_wh: Tuple[int, int], with_masks: bool
+    lines: List[str],
+    resolution_wh: Tuple[int, int],
+    with_masks: bool,
+    is_obb: bool = False,
 ) -> Detections:
    if len(lines) == 0:
        return Detections.empty()

-    class_id, relative_xyxy, relative_polygon = [], [], []
+    class_id, relative_xyxy, relative_polygon, relative_xyxyxyxy = [], [], [], []
    w, h = resolution_wh
    for line in lines:
        values = line.split()
@@ -88,21 +92,30 @@ def yolo_annotations_to_detections(
        elif len(values) > 5:
            polygon = _parse_polygon(values=values[1:])
            relative_xyxy.append(polygon_to_xyxy(polygon=polygon))
+            if is_obb:
+                relative_xyxyxyxy.append(np.array(values[1:]))
            if with_masks:
                relative_polygon.append(polygon)

    class_id = np.array(class_id, dtype=int)
    relative_xyxy = np.array(relative_xyxy, dtype=np.float32)
    xyxy = relative_xyxy * np.array([w, h, w, h], dtype=np.float32)
+    data = {}
+
+    if is_obb:
+        relative_xyxyxyxy = np.array(relative_xyxyxyxy, dtype=np.float32)
+        xyxyxyxy = relative_xyxyxyxy.reshape(-1, 4, 2)
+        xyxyxyxy *= np.array([w, h], dtype=np.float32)
+        data[ORIENTED_BOX_COORDINATES] = xyxyxyxy

    if not with_masks:
-        return Detections(class_id=class_id, xyxy=xyxy)
+        return Detections(class_id=class_id, xyxy=xyxy, data=data)

    polygons = [
        (polygon * np.array(resolution_wh)).astype(int) for polygon in relative_polygon
    ]
    mask = _polygons_to_masks(polygons=polygons, resolution_wh=resolution_wh)
-    return Detections(class_id=class_id, xyxy=xyxy, mask=mask)
+    return Detections(class_id=class_id, xyxy=xyxy, data=data, mask=mask)


 def load_yolo_annotations(
@@ -110,6 +123,7 @@ def load_yolo_annotations(
    annotations_directory_path: str,
    data_yaml_path: str,
    force_masks: bool = False,
+    is_obb: bool = False,
 ) -> Tuple[List[str], Dict[str, np.ndarray], Dict[str, Detections]]:
    """
    Loads YOLO annotations and returns class names, images,
@@ -123,6 +137,9 @@ def load_yolo_annotations(
            YAML file containing class information.
        force_masks (bool, optional): If True, forces masks to be loaded
            for all annotations, regardless of whether they are present.
+        is_obb (bool, optional): If True, loads the annotations in OBB format.
+            OBB annotations are defined as `[class_id, x, y, x, y, x, y, x, y]`,
+            where pairs of [x, y] are box corners.

    Returns:
        Tuple[List[str], Dict[str, np.ndarray], Dict[str, Detections]]:
@@ -156,9 +173,11 @@ def load_yolo_annotations(
        with_masks = _with_mask(lines=lines)
        with_masks = force_masks if force_masks else with_masks
        annotation = yolo_annotations_to_detections(
-            lines=lines, resolution_wh=resolution_wh, with_masks=with_masks
+            lines=lines,
+            resolution_wh=resolution_wh,
+            with_masks=with_masks,
+            is_obb=is_obb,
        )
-
        images[image_path] = image
        annotations[image_path] = annotation
    return classes, images, annotations
@@ -2,10 +2,11 @@ import copy
 import os
 import random
 from pathlib import Path
-from typing import Dict, List, Optional, Tuple, TypeVar
+from typing import Dict, List, Optional, Tuple, TypeVar, Union

 import cv2
 import numpy as np
+import numpy.typing as npt

 from supervision.detection.core import Detections
 from supervision.detection.utils import (
@@ -129,3 +130,123 @@ def train_test_split(

    split_index = int(len(data) * train_ratio)
    return data[:split_index], data[split_index:]
+
+
+def rle_to_mask(
+    rle: Union[npt.NDArray[np.int_], List[int]], resolution_wh: Tuple[int, int]
+) -> npt.NDArray[np.bool_]:
+    """
+    Converts run-length encoding (RLE) to a binary mask.
+
+    Args:
+        rle (Union[npt.NDArray[np.int_], List[int]]): The 1D RLE array, the format
+            used in the COCO dataset (column-wise encoding, values of an array with
+            even indices represent the number of pixels assigned as background,
+            values of an array with odd indices represent the number of pixels
+            assigned as foreground object).
+        resolution_wh (Tuple[int, int]): The width (w) and height (h)
+            of the desired binary mask.
+
+    Returns:
+        The generated 2D Boolean mask of shape `(h, w)`, where the foreground object is
+            marked with `True`'s and the rest is filled with `False`'s.
+
+    Raises:
+        AssertionError: If the sum of pixels encoded in RLE differs from the
+            number of pixels in the expected mask (computed based on resolution_wh).
+
+    Examples:
+        ```python
+        import supervision as sv
+
+        sv.rle_to_mask([5, 2, 2, 2, 5], (4, 4))
+        # array([
+        #     [False, False, False, False],
+        #     [False, True,  True,  False],
+        #     [False, True,  True,  False],
+        #     [False, False, False, False],
+        # ])
+        ```
+    """
+    if isinstance(rle, list):
+        rle = np.array(rle, dtype=int)
+
+    width, height = resolution_wh
+
+    assert width * height == np.sum(rle), (
+        "the sum of the number of pixels in the RLE must be the same "
+        "as the number of pixels in the expected mask"
+    )
+
+    zero_one_values = np.zeros(shape=(rle.size, 1), dtype=np.uint8)
+    zero_one_values[1::2] = 1
+
+    decoded_rle = np.repeat(zero_one_values, rle, axis=0)
+    decoded_rle = np.append(
+        decoded_rle, np.zeros(width * height - len(decoded_rle), dtype=np.uint8)
+    )
+    return decoded_rle.reshape((height, width), order="F")
+
+
+def mask_to_rle(mask: npt.NDArray[np.bool_]) -> List[int]:
+    """
+    Converts a binary mask into a run-length encoding (RLE).
+
+    Args:
+        mask (npt.NDArray[np.bool_]): 2D binary mask where `True` indicates foreground
+            object and `False` indicates background.
+
+    Returns:
+        The run-length encoded mask. Values of a list with even indices
+            represent the number of pixels assigned as background (`False`), values
+            of a list with odd indices represent the number of pixels assigned
+            as foreground object (`True`).
+
+    Raises:
+        AssertionError: If input mask is not 2D or is empty.
+
+    Examples:
+        ```python
+        import numpy as np
+        import supervision as sv
+
+        mask = np.array([
+            [True, True, True, True],
+            [True, True, True, True],
+            [True, True, True, True],
+            [True, True, True, True],
+        ])
+        sv.mask_to_rle(mask)
+        # [0, 16]
+
+        mask = np.array([
+            [False, False, False, False],
+            [False, True,  True,  False],
+            [False, True,  True,  False],
+            [False, False, False, False],
+        ])
+        sv.mask_to_rle(mask)
+        # [5, 2, 2, 2, 5]
+        ```
+
+    ![mask_to_rle](https://media.roboflow.com/supervision-docs/mask-to-rle.png){ align=center width="800" }
+    """  # noqa E501 // docs
+    assert mask.ndim == 2, "Input mask must be 2D"
+    assert mask.size != 0, "Input mask cannot be empty"
+
+    on_value_change_indices = np.where(
+        mask.ravel(order="F") != np.roll(mask.ravel(order="F"), 1)
+    )[0]
+
+    on_value_change_indices = np.append(on_value_change_indices, mask.size)
+    # need to add 0 at the beginning when the same value is in the first and
+    # last element of the flattened mask
+    if on_value_change_indices[0] != 0:
+        on_value_change_indices = np.insert(on_value_change_indices, 0, 0)
+
+    rle = np.diff(on_value_change_indices)
+
+    if mask[0][0] == 1:
+        rle = np.insert(rle, 0, 0)
+
+    return list(rle)
@@ -3,16 +3,12 @@ from typing import List, Optional, Union
 import cv2

 from supervision.annotators.base import ImageType
-from supervision.annotators.utils import scene_to_annotator_img_type
 from supervision.detection.core import Detections
 from supervision.draw.color import Color, ColorPalette
+from supervision.utils.conversion import ensure_cv2_image_for_annotation
 from supervision.utils.internal import deprecated


-@deprecated(
-    "`BoxAnnotator` is deprecated and will be removed in "
-    "`supervision-0.22.0`. Use `BoundingBoxAnnotator` and `LabelAnnotator` instead"
-)
 class BoxAnnotator:
    """
    A class for drawing bounding boxes on an image using detections provided.
@@ -46,7 +42,11 @@ class BoxAnnotator:
        self.text_thickness: int = text_thickness
        self.text_padding: int = text_padding

-    @scene_to_annotator_img_type
+    @deprecated(
+        "`BoxAnnotator` is deprecated and will be removed in "
+        "`supervision-0.22.0`. Use `BoundingBoxAnnotator` and `LabelAnnotator` instead"
+    )
+    @ensure_cv2_image_for_annotation
    def annotate(
        self,
        scene: ImageType,
@@ -7,53 +7,101 @@ from typing import Any, Dict, Iterator, List, Optional, Tuple, Union
 import numpy as np

 from supervision.config import CLASS_NAME_DATA_FIELD, ORIENTED_BOX_COORDINATES
-from supervision.detection.utils import (
+from supervision.detection.lmm import (
+    LMM,
+    from_florence_2,
+    from_paligemma,
+    validate_lmm_parameters,
+)
+from supervision.detection.overlap_filter import (
+    box_non_max_merge,
    box_non_max_suppression,
+    mask_non_max_suppression,
+)
+from supervision.detection.utils import (
+    box_iou_batch,
    calculate_masks_centroids,
    extract_ultralytics_masks,
    get_data_item,
    is_data_equal,
-    mask_non_max_suppression,
+    mask_to_xyxy,
    merge_data,
    process_roboflow_result,
-    validate_detections_fields,
    xywh_to_xyxy,
 )
 from supervision.geometry.core import Position
-from supervision.utils.internal import deprecated
+from supervision.utils.internal import deprecated, get_instance_variables
+from supervision.validators import validate_detections_fields


@dataclass
 class Detections:
    """
-    The `sv.Detections` allows you to convert results from a variety of object detection
-    and segmentation models into a single, unified format. The `sv.Detections` class
-    enables easy data manipulation and filtering, and provides a consistent API for
-    Supervision's tools like trackers, annotators, and zones.
+    The `sv.Detections` class in the Supervision library standardizes results from
+    various object detection and segmentation models into a consistent format. This
+    class simplifies data manipulation and filtering, providing a uniform API for
+    integration with Supervision [trackers](/trackers/), [annotators](/detection/annotators/), and [tools](/detection/tools/line_zone/).

-    ```python
-    import cv2
-    import supervision as sv
-    from ultralytics import YOLO
+    === "Inference"

-    image = cv2.imread(<SOURCE_IMAGE_PATH>)
-    model = YOLO('yolov8s.pt')
-    annotator = sv.BoundingBoxAnnotator()
+        Use [`sv.Detections.from_inference`](/detection/core/#supervision.detection.core.Detections.from_inference)
+        method, which accepts model results from both detection and segmentation models.

-    result = model(image)[0]
-    detections = sv.Detections.from_ultralytics(result)
+        ```python
+        import cv2
+        import supervision as sv
+        from inference import get_model

-    annotated_image = annotator.annotate(image, detections)
-    ```
+        model = get_model(model_id="yolov8n-640")
+        image = cv2.imread(<SOURCE_IMAGE_PATH>)
+        results = model.infer(image)[0]
+        detections = sv.Detections.from_inference(results)
+        ```

-    !!! tip
+    === "Ultralytics"

-        In `sv.Detections`, detection data is categorized into two main field types:
-        fixed and custom. The fixed fields include `xyxy`, `mask`, `confidence`,
-        `class_id`, and `tracker_id`. For any additional data requirements, custom
-        fields come into play, stored in the data field. These custom fields are easily
-        accessible using the `detections[<FIELD_NAME>]` syntax, providing flexibility
-        for diverse data handling needs.
+        Use [`sv.Detections.from_ultralytics`](/detection/core/#supervision.detection.core.Detections.from_ultralytics)
+        method, which accepts model results from both detection and segmentation models.
+
+        ```python
+        import cv2
+        import supervision as sv
+        from ultralytics import YOLO
+
+        model = YOLO("yolov8n.pt")
+        image = cv2.imread(<SOURCE_IMAGE_PATH>)
+        results = model(image)[0]
+        detections = sv.Detections.from_ultralytics(results)
+        ```
+
+    === "Transformers"
+
+        Use [`sv.Detections.from_transformers`](/detection/core/#supervision.detection.core.Detections.from_transformers)
+        method, which accepts model results from both detection and segmentation models.
+
+        ```python
+        import torch
+        import supervision as sv
+        from PIL import Image
+        from transformers import DetrImageProcessor, DetrForObjectDetection
+
+        processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+        model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+
+        image = Image.open(<SOURCE_IMAGE_PATH>)
+        inputs = processor(images=image, return_tensors="pt")
+
+        with torch.no_grad():
+            outputs = model(**inputs)
+
+        width, height = image.size
+        target_size = torch.tensor([[height, width]])
+        results = processor.post_process_object_detection(
+            outputs=outputs, target_sizes=target_size)[0]
+        detections = sv.Detections.from_transformers(
+            transformers_results=results,
+            id2label=model.config.id2label)
+        ```

    Attributes:
        xyxy (np.ndarray): An array of shape `(n, 4)` containing
@@ -69,15 +117,7 @@ class Detections:
        data (Dict[str, Union[np.ndarray, List]]): A dictionary containing additional
            data where each key is a string representing the data type, and the value
            is either a NumPy array or a list of corresponding data.
-
-    !!! warning
-
-        The `data` field in the `sv.Detections` class is currently in an experimental
-        phase. Please be aware that its API and functionality are subject to change in
-        future updates as we continue to refine and improve its capabilities.
-        We encourage users to experiment with this feature and provide feedback, but
-        also to be prepared for potential modifications in upcoming releases.
-    """
+    """  # noqa: E501 // docs

    xyxy: np.ndarray
    mask: Optional[np.ndarray] = None
@@ -176,8 +216,8 @@ class Detections:
    @classmethod
    def from_ultralytics(cls, ultralytics_results) -> Detections:
        """
-        Creates a Detections instance from a
-            [YOLOv8](https://github.com/ultralytics/ultralytics) inference result.
+        Creates a `sv.Detections` instance from a
+        [YOLOv8](https://github.com/ultralytics/ultralytics) inference result.

        !!! Note

@@ -188,7 +228,7 @@ class Detections:

        Args:
            ultralytics_results (ultralytics.yolo.engine.results.Results):
-                The output Results instance from YOLOv8
+                The output Results instance from Ultralytics

        Returns:
            Detections: A new Detections object.
@@ -201,13 +241,16 @@ class Detections:

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
            model = YOLO('yolov8s.pt')
-
-            result = model(image)[0]
-            detections = sv.Detections.from_ultralytics(result)
+            results = model(image)[0]
+            detections = sv.Detections.from_ultralytics(results)
            ```
+
+        !!! tip
+
+            Class names values can be accessed using `detections["class_name"]`.
        """  # noqa: E501 // docs

-        if ultralytics_results.obb is not None:
+        if hasattr(ultralytics_results, "obb") and ultralytics_results.obb is not None:
            class_id = ultralytics_results.obb.cls.cpu().numpy().astype(int)
            class_names = np.array([ultralytics_results.names[i] for i in class_id])
            oriented_box_coordinates = ultralytics_results.obb.xyxyxyxy.cpu().numpy()
@@ -385,23 +428,86 @@ class Detections:
            xyxy=mmdet_results.pred_instances.bboxes.cpu().numpy(),
            confidence=mmdet_results.pred_instances.scores.cpu().numpy(),
            class_id=mmdet_results.pred_instances.labels.cpu().numpy().astype(int),
+            mask=mmdet_results.pred_instances.masks.cpu().numpy()
+            if "masks" in mmdet_results.pred_instances
+            else None,
        )

    @classmethod
-    def from_transformers(cls, transformers_results: dict) -> Detections:
+    def from_transformers(
+        cls, transformers_results: dict, id2label: Optional[Dict[int, str]] = None
+    ) -> Detections:
        """
-        Creates a Detections instance from object detection
-        [transformer](https://github.com/huggingface/transformers) inference result.
+        Creates a Detections instance from object detection or segmentation
+        [Transformer](https://github.com/huggingface/transformers) inference result.
+
+        Args:
+            transformers_results (dict): The output of Transformers model inference. A
+                dictionary containing the `scores`, `labels`, `boxes` and `masks` keys.
+            id2label (Optional[Dict[int, str]]): A dictionary mapping class IDs to
+                class names. If provided, the resulting Detections object will contain
+                `class_name` data field with the class names.

        Returns:
            Detections: A new Detections object.
-        """

-        return cls(
-            xyxy=transformers_results["boxes"].cpu().numpy(),
-            confidence=transformers_results["scores"].cpu().numpy(),
-            class_id=transformers_results["labels"].cpu().numpy().astype(int),
-        )
+        Example:
+            ```python
+            import torch
+            import supervision as sv
+            from PIL import Image
+            from transformers import DetrImageProcessor, DetrForObjectDetection
+
+            processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
+            model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
+
+            image = Image.open(<SOURCE_IMAGE_PATH>)
+            inputs = processor(images=image, return_tensors="pt")
+
+            with torch.no_grad():
+                outputs = model(**inputs)
+
+            width, height = image.size
+            target_size = torch.tensor([[height, width]])
+            results = processor.post_process_object_detection(
+                outputs=outputs, target_sizes=target_size)[0]
+
+            detections = sv.Detections.from_transformers(
+                transformers_results=results,
+                id2label=model.config.id2label
+            )
+            ```
+
+        !!! tip
+
+            Class names values can be accessed using `detections["class_name"]`.
+        """  # noqa: E501 // docs
+
+        class_ids = transformers_results["labels"].cpu().detach().numpy().astype(int)
+        data = {}
+        if id2label is not None:
+            class_names = np.array([id2label[class_id] for class_id in class_ids])
+            data[CLASS_NAME_DATA_FIELD] = class_names
+        if "boxes" in transformers_results:
+            return cls(
+                xyxy=transformers_results["boxes"].cpu().detach().numpy(),
+                confidence=transformers_results["scores"].cpu().detach().numpy(),
+                class_id=class_ids,
+                data=data,
+            )
+        elif "masks" in transformers_results:
+            masks = transformers_results["masks"].cpu().detach().numpy().astype(bool)
+            return cls(
+                xyxy=mask_to_xyxy(masks),
+                mask=masks,
+                confidence=transformers_results["scores"].cpu().detach().numpy(),
+                class_id=class_ids,
+                data=data,
+            )
+        else:
+            raise NotImplementedError(
+                "Only object detection and semantic segmentation results are supported."
+            )

    @classmethod
    def from_detectron2(cls, detectron2_results) -> Detections:
@@ -439,6 +545,9 @@ class Detections:
        return cls(
            xyxy=detectron2_results["instances"].pred_boxes.tensor.cpu().numpy(),
            confidence=detectron2_results["instances"].scores.cpu().numpy(),
+            mask=detectron2_results["instances"].pred_masks.cpu().numpy()
+            if hasattr(detectron2_results["instances"], "pred_masks")
+            else None,
            class_id=detectron2_results["instances"]
            .pred_classes.cpu()
            .numpy()
@@ -448,17 +557,12 @@ class Detections:
    @classmethod
    def from_inference(cls, roboflow_result: Union[dict, Any]) -> Detections:
        """
-        Create a Detections object from the [Roboflow](https://roboflow.com/)
+        Create a `sv.Detections` object from the [Roboflow](https://roboflow.com/)
        API inference result or the [Inference](https://inference.roboflow.com/)
        package results. This method extracts bounding boxes, class IDs,
        confidences, and class names from the Roboflow API result and encapsulates
        them into a Detections object.

-        !!! note
-
-            Class names can be accessed using the key 'class_name' in the returned
-            object's data attribute.
-
        Args:
            roboflow_result (dict, any): The result from the
                Roboflow API or Inference package containing predictions.
@@ -471,14 +575,18 @@ class Detections:
            ```python
            import cv2
            import supervision as sv
-            from inference.models.utils import get_roboflow_model
+            from inference import get_model

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
-            model = get_roboflow_model(model_id="yolov8s-640")
+            model = get_model(model_id="yolov8s-640")

            result = model.infer(image)[0]
            detections = sv.Detections.from_inference(result)
            ```
+
+        !!! tip
+
+            Class names values can be accessed using `detections["class_name"]`.
        """
        with suppress(AttributeError):
            roboflow_result = roboflow_result.dict(exclude_none=True, by_alias=True)
@@ -528,10 +636,10 @@ class Detections:
            ```python
            import cv2
            import supervision as sv
-            from inference.models.utils import get_roboflow_model
+            from inference import get_model

            image = cv2.imread(<SOURCE_IMAGE_PATH>)
-            model = get_roboflow_model(model_id="yolov8s-640")
+            model = get_model(model_id="yolov8s-640")

            result = model.infer(image)[0]
            detections = sv.Detections.from_roboflow(result)
@@ -710,6 +818,69 @@ class Detections:
            class_id=paddledet_result["bbox"][:, 0].astype(int),
        )

+    @classmethod
+    def from_lmm(
+        cls, lmm: Union[LMM, str], result: Union[str, dict], **kwargs
+    ) -> Detections:
+        """
+        Creates a Detections object from the given result string based on the specified
+        Large Multimodal Model (LMM).
+
+        Args:
+            lmm (Union[LMM, str]): The type of LMM (Large Multimodal Model) to use.
+            result (str): The result string containing the detection data.
+            **kwargs: Additional keyword arguments required by the specified LMM.
+
+        Returns:
+            Detections: A new Detections object.
+
+        Raises:
+            ValueError: If the LMM is invalid, required arguments are missing, or
+                disallowed arguments are provided.
+            ValueError: If the specified LMM is not supported.
+
+        Examples:
+            ```python
+            import supervision as sv
+
+            paligemma_result = "<loc0256><loc0256><loc0768><loc0768> cat"
+            detections = sv.Detections.from_lmm(
+                sv.LMM.PALIGEMMA,
+                paligemma_result,
+                resolution_wh=(1000, 1000),
+                classes=['cat', 'dog']
+            )
+            detections.xyxy
+            # array([[250., 250., 750., 750.]])
+
+            detections.class_id
+            # array([0])
+            ```
+        """
+        lmm = validate_lmm_parameters(lmm, result, kwargs)
+
+        if lmm == LMM.PALIGEMMA:
+            assert isinstance(result, str)
+            xyxy, class_id, class_name = from_paligemma(result, **kwargs)
+            data = {CLASS_NAME_DATA_FIELD: class_name}
+            return cls(xyxy=xyxy, class_id=class_id, data=data)
+
+        if lmm == LMM.FLORENCE_2:
+            assert isinstance(result, dict)
+            xyxy, labels, mask, xyxyxyxy = from_florence_2(result, **kwargs)
+            if len(xyxy) == 0:
+                return cls.empty()
+
+            data = {}
+            if labels is not None:
+                data[CLASS_NAME_DATA_FIELD] = labels
+            if xyxyxyxy is not None:
+                data[ORIENTED_BOX_COORDINATES] = xyxyxyxy
+
+            return cls(xyxy=xyxy, mask=mask, data=data)
+
+        raise ValueError(f"Unsupported LMM: {lmm}")
+
    @classmethod
    def empty(cls) -> Detections:
        """
@@ -732,6 +903,14 @@ class Detections:
            class_id=np.array([], dtype=int),
        )

+    def is_empty(self) -> bool:
+        """
+        Returns `True` if the `Detections` object is considered empty.
+        """
+        empty_detections = Detections.empty()
+        empty_detections.data = self.data
+        return self == empty_detections
+
    @classmethod
    def merge(cls, detections_list: List[Detections]) -> Detections:
        """
@@ -739,9 +918,14 @@ class Detections:

        This method takes a list of Detections objects and combines their
        respective fields (`xyxy`, `mask`, `confidence`, `class_id`, and `tracker_id`)
-        into a single Detections object. If all elements in a field are not
-        `None`, the corresponding field will be stacked.
-        Otherwise, the field will be set to `None`.
+        into a single Detections object.
+
+        For example, if merging Detections with 3 and 4 detected objects, this method
+        will return a Detections with 7 objects (7 entries in `xyxy`, `mask`, etc).
+
+        !!! Note
+
+            When merging, empty `Detections` objects are ignored.

        Args:
            detections_list (List[Detections]): A list of Detections objects to merge.
@@ -781,6 +965,10 @@ class Detections:
            array([0.1, 0.2, 0.3])
            ```
        """
+        detections_list = [
+            detections for detections in detections_list if not detections.is_empty()
+        ]
+
        if len(detections_list) == 0:
            return Detections.empty()

@@ -1055,3 +1243,195 @@ class Detections:
            )

        return self[indices]
+
+    def with_nmm(
+        self, threshold: float = 0.5, class_agnostic: bool = False
+    ) -> Detections:
+        """
+        Perform non-maximum merging on the current set of object detections.
+
+        Args:
+            threshold (float, optional): The intersection-over-union threshold
+                to use for non-maximum merging. Defaults to 0.5.
+            class_agnostic (bool, optional): Whether to perform class-agnostic
+                non-maximum merging. If True, the class_id of each detection
+                will be ignored. Defaults to False.
+
+        Returns:
+            Detections: A new Detections object containing the subset of detections
+                after non-maximum merging.
+
+        Raises:
+            AssertionError: If `confidence` is None or `class_id` is None and
+                class_agnostic is False.
+
+        ![non-max-merging](https://media.roboflow.com/supervision-docs/non-max-merging.png){ align=center width="800" }
+        """  # noqa: E501 // docs
+        if len(self) == 0:
+            return self
+
+        assert (
+            self.confidence is not None
+        ), "Detections confidence must be given for NMM to be executed."
+
+        if class_agnostic:
+            predictions = np.hstack((self.xyxy, self.confidence.reshape(-1, 1)))
+        else:
+            assert self.class_id is not None, (
+                "Detections class_id must be given for NMM to be executed. If you"
+                " intended to perform class agnostic NMM set class_agnostic=True."
+            )
+            predictions = np.hstack(
+                (
+                    self.xyxy,
+                    self.confidence.reshape(-1, 1),
+                    self.class_id.reshape(-1, 1),
+                )
+            )
+
+        merge_groups = box_non_max_merge(
+            predictions=predictions, iou_threshold=threshold
+        )
+
+        result = []
+        for merge_group in merge_groups:
+            unmerged_detections = [self[i] for i in merge_group]
+            merged_detections = merge_inner_detections_objects(
+                unmerged_detections, threshold
+            )
+            result.append(merged_detections)
+
+        return Detections.merge(result)
+
+
+def merge_inner_detection_object_pair(
+    detections_1: Detections, detections_2: Detections
+) -> Detections:
+    """
+    Merges two Detections object into a single Detections object.
+    Assumes each Detections contains exactly one object.
+
+    A `winning` detection is determined based on the confidence score of the two
+    input detections. This winning detection is then used to specify which
+    `class_id`, `tracker_id`, and `data` to include in the merged Detections object.
+
+    The resulting `confidence` of the merged object is calculated by the weighted
+    contribution of ea detection to the merged object.
+    The bounding boxes and masks of the two input detections are merged into a
+    single bounding box and mask, respectively.
+
+    Args:
+        detections_1 (Detections):
+            The first Detections object
+        detections_2 (Detections):
+            The second Detections object
+
+    Returns:
+        Detections: A new Detections object, with merged attributes.
+
+    Raises:
+        ValueError: If the input Detections objects do not have exactly 1 detected
+            object.
+
+    Example:
+        ```python
+        import cv2
+        import supervision as sv
+        from inference import get_model
+
+        image = cv2.imread(<SOURCE_IMAGE_PATH>)
+        model = get_model(model_id="yolov8s-640")
+
+        result = model.infer(image)[0]
+        detections = sv.Detections.from_inference(result)
+
+        merged_detections = merge_object_detection_pair(
+            detections[0], detections[1])
+        ```
+    """
+    if len(detections_1) != 1 or len(detections_2) != 1:
+        raise ValueError("Both Detections should have exactly 1 detected object.")
+
+    validate_fields_both_defined_or_none(detections_1, detections_2)
+
+    xyxy_1 = detections_1.xyxy[0]
+    xyxy_2 = detections_2.xyxy[0]
+    if detections_1.confidence is None and detections_2.confidence is None:
+        merged_confidence = None
+    else:
+        detection_1_area = (xyxy_1[2] - xyxy_1[0]) * (xyxy_1[3] - xyxy_1[1])
+        detections_2_area = (xyxy_2[2] - xyxy_2[0]) * (xyxy_2[3] - xyxy_2[1])
+        merged_confidence = (
+            detection_1_area * detections_1.confidence[0]
+            + detections_2_area * detections_2.confidence[0]
+        ) / (detection_1_area + detections_2_area)
+        merged_confidence = np.array([merged_confidence])
+
+    merged_x1, merged_y1 = np.minimum(xyxy_1[:2], xyxy_2[:2])
+    merged_x2, merged_y2 = np.maximum(xyxy_1[2:], xyxy_2[2:])
+    merged_xyxy = np.array([[merged_x1, merged_y1, merged_x2, merged_y2]])
+
+    if detections_1.mask is None and detections_2.mask is None:
+        merged_mask = None
+    else:
+        merged_mask = np.logical_or(detections_1.mask, detections_2.mask)
+
+    if detections_1.confidence is None and detections_2.confidence is None:
+        winning_detection = detections_1
+    elif detections_1.confidence[0] >= detections_2.confidence[0]:
+        winning_detection = detections_1
+    else:
+        winning_detection = detections_2
+
+    return Detections(
+        xyxy=merged_xyxy,
+        mask=merged_mask,
+        confidence=merged_confidence,
+        class_id=winning_detection.class_id,
+        tracker_id=winning_detection.tracker_id,
+        data=winning_detection.data,
+    )
+
+
+def merge_inner_detections_objects(
+    detections: List[Detections], threshold=0.5
+) -> Detections:
+    """
+    Given N detections each of length 1 (exactly one object inside), combine them into a
+    single detection object of length 1. The contained inner object will be the merged
+    result of all the input detections.
+
+    For example, this lets you merge N boxes into one big box, N masks into one mask,
+    etc.
+    """
+    detections_1 = detections[0]
+    for detections_2 in detections[1:]:
+        box_iou = box_iou_batch(detections_1.xyxy, detections_2.xyxy)[0]
+        if box_iou < threshold:
+            break
+        detections_1 = merge_inner_detection_object_pair(detections_1, detections_2)
+    return detections_1
+
+
+def validate_fields_both_defined_or_none(
+    detections_1: Detections, detections_2: Detections
+) -> None:
+    """
+    Verify that for each optional field in the Detections, both instances either have
+    the field set to None or both have it set to non-None values.
+
+    `data` field is ignored.
+
+    Raises:
+        ValueError: If one field is None and the other is not, for any of the fields.
+    """
+    attributes = get_instance_variables(detections_1)
+    for attribute in attributes:
+        value_1 = getattr(detections_1, attribute)
+        value_2 = getattr(detections_2, attribute)
+
+        if (value_1 is None) != (value_2 is None):
+            raise ValueError(
+                f"Field '{attribute}' should be consistently None or not None in both "
+                "Detections."
+            )
@@ -1,12 +1,15 @@
+import warnings
 from typing import Dict, Iterable, Optional, Tuple

 import cv2
 import numpy as np

 from supervision.detection.core import Detections
+from supervision.detection.utils import cross_product
 from supervision.draw.color import Color
 from supervision.draw.utils import draw_text
 from supervision.geometry.core import Point, Position, Vector
+from supervision.utils.internal import SupervisionWarnings


 class LineZone:
@@ -81,6 +84,8 @@ class LineZone:
        self.in_count: int = 0
        self.out_count: int = 0
        self.triggering_anchors = triggering_anchors
+        if not list(self.triggering_anchors):
+            raise ValueError("Triggering anchors cannot be empty.")

    @staticmethod
    def calculate_region_of_interest_limits(vector: Vector) -> Tuple[Vector, Vector]:
@@ -140,6 +145,15 @@ class LineZone:
        if len(detections) == 0:
            return crossed_in, crossed_out

+        if detections.tracker_id is None:
+            warnings.warn(
+                "Line zone counting skipped. LineZone requires tracker_id. Refer to "
+                "https://supervision.roboflow.com/latest/trackers for more "
+                "information.",
+                category=SupervisionWarnings,
+            )
+            return crossed_in, crossed_out
+
        all_anchors = np.array(
            [
                detections.get_anchors_coordinates(anchor)
@@ -147,31 +161,23 @@ class LineZone:
            ]
        )

+        cross_products_1 = cross_product(all_anchors, self.limits[0])
+        cross_products_2 = cross_product(all_anchors, self.limits[1])
+        in_limits = (cross_products_1 > 0) == (cross_products_2 > 0)
+        in_limits = np.all(in_limits, axis=0)
+
+        triggers = cross_product(all_anchors, self.vector) < 0
+        has_any_left_trigger = np.any(triggers, axis=0)
+        has_any_right_trigger = np.any(~triggers, axis=0)
+        is_uniformly_triggered = ~(has_any_left_trigger & has_any_right_trigger)
        for i, tracker_id in enumerate(detections.tracker_id):
-            if tracker_id is None:
+            if not in_limits[i]:
                continue

-            box_anchors = [Point(x=x, y=y) for x, y in all_anchors[:, i, :]]
-
-            in_limits = all(
-                [
-                    self.is_point_in_limits(point=anchor, limits=self.limits)
-                    for anchor in box_anchors
-                ]
-            )
-
-            if not in_limits:
+            if not is_uniformly_triggered[i]:
                continue

-            triggers = [
-                self.vector.cross_product(point=anchor) < 0 for anchor in box_anchors
-            ]
-
-            if len(set(triggers)) == 2:
-                continue
-
-            tracker_state = triggers[0]
-
+            tracker_state = has_any_left_trigger[i]
            if tracker_id not in self.tracker_state:
                self.tracker_state[tracker_id] = tracker_state
                continue
@@ -0,0 +1,182 @@
+import re
+from enum import Enum
+from typing import Any, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+
+from supervision.detection.utils import polygon_to_mask, polygon_to_xyxy
+
+
+class LMM(Enum):
+    PALIGEMMA = "paligemma"
+    FLORENCE_2 = "florence_2"
+
+
+RESULT_TYPES: Dict[LMM, type] = {LMM.PALIGEMMA: str, LMM.FLORENCE_2: dict}
+
+REQUIRED_ARGUMENTS: Dict[LMM, List[str]] = {
+    LMM.PALIGEMMA: ["resolution_wh"],
+    LMM.FLORENCE_2: ["resolution_wh"],
+}
+
+ALLOWED_ARGUMENTS: Dict[LMM, List[str]] = {
+    LMM.PALIGEMMA: ["resolution_wh", "classes"],
+    LMM.FLORENCE_2: ["resolution_wh"],
+}
+
+SUPPORTED_TASKS_FLORENCE_2 = [
+    "<OD>",
+    "<CAPTION_TO_PHRASE_GROUNDING>",
+    "<DENSE_REGION_CAPTION>",
+    "<REGION_PROPOSAL>",
+    "<OCR_WITH_REGION>",
+    "<REFERRING_EXPRESSION_SEGMENTATION>",
+    "<REGION_TO_SEGMENTATION>",
+    "<OPEN_VOCABULARY_DETECTION>",
+    "<REGION_TO_CATEGORY>",
+    "<REGION_TO_DESCRIPTION>",
+]
+
+
+def validate_lmm_parameters(
+    lmm: Union[LMM, str], result: Any, kwargs: Dict[str, Any]
+) -> LMM:
+    if isinstance(lmm, str):
+        try:
+            lmm = LMM(lmm.lower())
+        except ValueError:
+            raise ValueError(
+                f"Invalid lmm value: {lmm}. Must be one of {[e.value for e in LMM]}"
+            )
+
+    if not isinstance(result, RESULT_TYPES[lmm]):
+        raise ValueError(
+            f"Invalid LMM result type: {type(result)}. Must be {RESULT_TYPES[lmm]}"
+        )
+
+    required_args = REQUIRED_ARGUMENTS.get(lmm, [])
+    for arg in required_args:
+        if arg not in kwargs:
+            raise ValueError(f"Missing required argument: {arg}")
+
+    allowed_args = ALLOWED_ARGUMENTS.get(lmm, [])
+    for arg in kwargs:
+        if arg not in allowed_args:
+            raise ValueError(f"Argument {arg} is not allowed for {lmm.name}")
+
+    return lmm
+
+
+def from_paligemma(
+    result: str, resolution_wh: Tuple[int, int], classes: Optional[List[str]] = None
+) -> Tuple[np.ndarray, Optional[np.ndarray], np.ndarray]:
+    w, h = resolution_wh
+    pattern = re.compile(
+        r"(?<!<loc\d{4}>)<loc(\d{4})><loc(\d{4})><loc(\d{4})><loc(\d{4})> ([\w\s\-]+)"
+    )
+    matches = pattern.findall(result)
+    matches = np.array(matches) if matches else np.empty((0, 5))
+
+    xyxy, class_name = matches[:, [1, 0, 3, 2]], matches[:, 4]
+    xyxy = xyxy.astype(int) / 1024 * np.array([w, h, w, h])
+    class_name = np.char.strip(class_name.astype(str))
+    class_id = None
+
+    if classes is not None:
+        mask = np.array([name in classes for name in class_name]).astype(bool)
+        xyxy, class_name = xyxy[mask], class_name[mask]
+        class_id = np.array([classes.index(name) for name in class_name])
+
+    return xyxy, class_id, class_name
+
+
+def from_florence_2(
+    result: dict, resolution_wh: Tuple[int, int]
+) -> Tuple[
+    np.ndarray, Optional[np.ndarray], Optional[np.ndarray], Optional[np.ndarray]
+]:
+    """
+    Parse results from the Florence 2 multi-model model.
+    https://huggingface.co/microsoft/Florence-2-large
+
+    Parameters:
+        result: dict containing the model output
+
+    Returns:
+        xyxy (np.ndarray): An array of shape `(n, 4)` containing
+            the bounding boxes coordinates in format `[x1, y1, x2, y2]`
+        labels: (Optional[np.ndarray]): An array of shape `(n,)` containing
+            the class labels for each bounding box
+        masks: (Optional[np.ndarray]): An array of shape `(n, h, w)` containing
+            the segmentation masks for each bounding box
+        obb_boxes: (Optional[np.ndarray]): An array of shape `(n, 4, 2)` containing
+            oriented bounding boxes.
+    """
+    assert len(result) == 1, f"Expected result with a single element. Got: {result}"
+    task = list(result.keys())[0]
+    if task not in SUPPORTED_TASKS_FLORENCE_2:
+        raise ValueError(
+            f"{task} not supported. Supported tasks are: {SUPPORTED_TASKS_FLORENCE_2}"
+        )
+    result = result[task]
+
+    if task in ["<OD>", "<CAPTION_TO_PHRASE_GROUNDING>", "<DENSE_REGION_CAPTION>"]:
+        xyxy = np.array(result["bboxes"], dtype=np.float32)
+        labels = np.array(result["labels"])
+        return xyxy, labels, None, None
+
+    if task == "<REGION_PROPOSAL>":
+        xyxy = np.array(result["bboxes"], dtype=np.float32)
+        # provides labels, but they are ["", "", "", ...]
+        return xyxy, None, None, None
+
+    if task == "<OCR_WITH_REGION>":
+        xyxyxyxy = np.array(result["quad_boxes"], dtype=np.float32)
+        xyxyxyxy = xyxyxyxy.reshape(-1, 4, 2)
+        xyxy = np.array([polygon_to_xyxy(polygon) for polygon in xyxyxyxy])
+        labels = np.array(result["labels"])
+        return xyxy, labels, None, xyxyxyxy
+
+    if task in ["<REFERRING_EXPRESSION_SEGMENTATION>", "<REGION_TO_SEGMENTATION>"]:
+        xyxy_list = []
+        masks_list = []
+        for polygons_of_same_class in result["polygons"]:
+            for polygon in polygons_of_same_class:
+                polygon = np.reshape(polygon, (-1, 2)).astype(np.int32)
+                mask = polygon_to_mask(polygon, resolution_wh).astype(bool)
+                masks_list.append(mask)
+                xyxy = polygon_to_xyxy(polygon)
+                xyxy_list.append(xyxy)
+            # per-class labels also provided, but they are ["", "", "", ...]
+            # when we figure out how to set class names, we can do
+            # zip(result["labels"], result["polygons"])
+        xyxy = np.array(xyxy_list, dtype=np.float32)
+        masks = np.array(masks_list)
+        return xyxy, None, masks, None
+
+    if task == "<OPEN_VOCABULARY_DETECTION>":
+        xyxy = np.array(result["bboxes"], dtype=np.float32)
+        labels = np.array(result["bboxes_labels"])
+        # Also has "polygons" and "polygons_labels", but they don't seem to be used
+        return xyxy, labels, None, None
+
+    if task in ["<REGION_TO_CATEGORY>", "<REGION_TO_DESCRIPTION>"]:
+        assert isinstance(
+            result, str
+        ), f"Expected string as <REGION_TO_CATEGORY> result, got {type(result)}"
+
+        if result == "No object detected.":
+            return np.empty((0, 4), dtype=np.float32), np.array([]), None, None
+
+        pattern = re.compile(r"<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>")
+        match = pattern.search(result)
+        assert (
+            match is not None
+        ), f"Expected string to end in location tags, but got {result}"
+
+        xyxy = np.array([match.groups()], dtype=np.float32)
+        result_string = result[: match.start()]
+        labels = np.array([result_string])
+        return xyxy, labels, None, None
+
+    assert False, f"Unimplemented task: {task}"
@@ -0,0 +1,263 @@
+from enum import Enum
+from typing import List, Union
+
+import numpy as np
+import numpy.typing as npt
+
+from supervision.detection.utils import box_iou_batch, mask_iou_batch
+
+
+def resize_masks(masks: np.ndarray, max_dimension: int = 640) -> np.ndarray:
+    """
+    Resize all masks in the array to have a maximum dimension of max_dimension,
+    maintaining aspect ratio.
+
+    Args:
+        masks (np.ndarray): 3D array of binary masks with shape (N, H, W).
+        max_dimension (int): The maximum dimension for the resized masks.
+
+    Returns:
+        np.ndarray: Array of resized masks.
+    """
+    max_height = np.max(masks.shape[1])
+    max_width = np.max(masks.shape[2])
+    scale = min(max_dimension / max_height, max_dimension / max_width)
+
+    new_height = int(scale * max_height)
+    new_width = int(scale * max_width)
+
+    x = np.linspace(0, max_width - 1, new_width).astype(int)
+    y = np.linspace(0, max_height - 1, new_height).astype(int)
+    xv, yv = np.meshgrid(x, y)
+
+    resized_masks = masks[:, yv, xv]
+
+    resized_masks = resized_masks.reshape(masks.shape[0], new_height, new_width)
+    return resized_masks
+
+
+def mask_non_max_suppression(
+    predictions: np.ndarray,
+    masks: np.ndarray,
+    iou_threshold: float = 0.5,
+    mask_dimension: int = 640,
+) -> np.ndarray:
+    """
+    Perform Non-Maximum Suppression (NMS) on segmentation predictions.
+
+    Args:
+        predictions (np.ndarray): A 2D array of object detection predictions in
+            the format of `(x_min, y_min, x_max, y_max, score)`
+            or `(x_min, y_min, x_max, y_max, score, class)`. Shape: `(N, 5)` or
+            `(N, 6)`, where N is the number of predictions.
+        masks (np.ndarray): A 3D array of binary masks corresponding to the predictions.
+            Shape: `(N, H, W)`, where N is the number of predictions, and H, W are the
+            dimensions of each mask.
+        iou_threshold (float, optional): The intersection-over-union threshold
+            to use for non-maximum suppression.
+        mask_dimension (int, optional): The dimension to which the masks should be
+            resized before computing IOU values. Defaults to 640.
+
+    Returns:
+        np.ndarray: A boolean array indicating which predictions to keep after
+            non-maximum suppression.
+
+    Raises:
+        AssertionError: If `iou_threshold` is not within the closed
+        range from `0` to `1`.
+    """
+    assert 0 <= iou_threshold <= 1, (
+        "Value of `iou_threshold` must be in the closed range from 0 to 1, "
+        f"{iou_threshold} given."
+    )
+    rows, columns = predictions.shape
+
+    if columns == 5:
+        predictions = np.c_[predictions, np.zeros(rows)]
+
+    sort_index = predictions[:, 4].argsort()[::-1]
+    predictions = predictions[sort_index]
+    masks = masks[sort_index]
+    masks_resized = resize_masks(masks, mask_dimension)
+    ious = mask_iou_batch(masks_resized, masks_resized)
+    categories = predictions[:, 5]
+
+    keep = np.ones(rows, dtype=bool)
+    for i in range(rows):
+        if keep[i]:
+            condition = (ious[i] > iou_threshold) & (categories[i] == categories)
+            keep[i + 1 :] = np.where(condition[i + 1 :], False, keep[i + 1 :])
+
+    return keep[sort_index.argsort()]
+
+
+def box_non_max_suppression(
+    predictions: np.ndarray, iou_threshold: float = 0.5
+) -> np.ndarray:
+    """
+    Perform Non-Maximum Suppression (NMS) on object detection predictions.
+
+    Args:
+        predictions (np.ndarray): An array of object detection predictions in
+            the format of `(x_min, y_min, x_max, y_max, score)`
+            or `(x_min, y_min, x_max, y_max, score, class)`.
+        iou_threshold (float, optional): The intersection-over-union threshold
+            to use for non-maximum suppression.
+
+    Returns:
+        np.ndarray: A boolean array indicating which predictions to keep after n
+            on-maximum suppression.
+
+    Raises:
+        AssertionError: If `iou_threshold` is not within the
+            closed range from `0` to `1`.
+    """
+    assert 0 <= iou_threshold <= 1, (
+        "Value of `iou_threshold` must be in the closed range from 0 to 1, "
+        f"{iou_threshold} given."
+    )
+    rows, columns = predictions.shape
+
+    # add column #5 - category filled with zeros for agnostic nms
+    if columns == 5:
+        predictions = np.c_[predictions, np.zeros(rows)]
+
+    # sort predictions column #4 - score
+    sort_index = np.flip(predictions[:, 4].argsort())
+    predictions = predictions[sort_index]
+
+    boxes = predictions[:, :4]
+    categories = predictions[:, 5]
+    ious = box_iou_batch(boxes, boxes)
+    ious = ious - np.eye(rows)
+
+    keep = np.ones(rows, dtype=bool)
+
+    for index, (iou, category) in enumerate(zip(ious, categories)):
+        if not keep[index]:
+            continue
+
+        # drop detections with iou > iou_threshold and
+        # same category as current detections
+        condition = (iou > iou_threshold) & (categories == category)
+        keep = keep & ~condition
+
+    return keep[sort_index.argsort()]
+
+
+def group_overlapping_boxes(
+    predictions: npt.NDArray[np.float64], iou_threshold: float = 0.5
+) -> List[List[int]]:
+    """
+    Apply greedy version of non-maximum merging to avoid detecting too many
+    overlapping bounding boxes for a given object.
+
+    Args:
+        predictions (npt.NDArray[np.float64]): An array of shape `(n, 5)` containing
+            the bounding boxes coordinates in format `[x1, y1, x2, y2]`
+            and the confidence scores.
+        iou_threshold (float, optional): The intersection-over-union threshold
+            to use for non-maximum suppression. Defaults to 0.5.
+
+    Returns:
+        List[List[int]]: Groups of prediction indices be merged.
+            Each group may have 1 or more elements.
+    """
+    merge_groups: List[List[int]] = []
+
+    scores = predictions[:, 4]
+    order = scores.argsort()
+
+    while len(order) > 0:
+        idx = int(order[-1])
+
+        order = order[:-1]
+        if len(order) == 0:
+            merge_groups.append([idx])
+            break
+
+        merge_candidate = np.expand_dims(predictions[idx], axis=0)
+        ious = box_iou_batch(predictions[order][:, :4], merge_candidate[:, :4])
+        ious = ious.flatten()
+
+        above_threshold = ious >= iou_threshold
+        merge_group = [idx] + np.flip(order[above_threshold]).tolist()
+        merge_groups.append(merge_group)
+        order = order[~above_threshold]
+    return merge_groups
+
+
+def box_non_max_merge(
+    predictions: npt.NDArray[np.float64],
+    iou_threshold: float = 0.5,
+) -> List[List[int]]:
+    """
+    Apply greedy version of non-maximum merging per category to avoid detecting
+    too many overlapping bounding boxes for a given object.
+
+    Args:
+        predictions (npt.NDArray[np.float64]): An array of shape `(n, 5)` or `(n, 6)`
+            containing the bounding boxes coordinates in format `[x1, y1, x2, y2]`,
+            the confidence scores and class_ids. Omit class_id column to allow
+            detections of different classes to be merged.
+        iou_threshold (float, optional): The intersection-over-union threshold
+            to use for non-maximum suppression. Defaults to 0.5.
+
+    Returns:
+        List[List[int]]: Groups of prediction indices be merged.
+            Each group may have 1 or more elements.
+    """
+    if predictions.shape[1] == 5:
+        return group_overlapping_boxes(predictions, iou_threshold)
+
+    category_ids = predictions[:, 5]
+    merge_groups = []
+    for category_id in np.unique(category_ids):
+        curr_indices = np.where(category_ids == category_id)[0]
+        merge_class_groups = group_overlapping_boxes(
+            predictions[curr_indices], iou_threshold
+        )
+
+        for merge_class_group in merge_class_groups:
+            merge_groups.append(curr_indices[merge_class_group].tolist())
+
+    for merge_group in merge_groups:
+        if len(merge_group) == 0:
+            raise ValueError(
+                f"Empty group detected when non-max-merging "
+                f"detections: {merge_groups}"
+            )
+    return merge_groups
+
+
+class OverlapFilter(Enum):
+    """
+    Enum specifying the strategy for filtering overlapping detections.
+
+    Attributes:
+        NONE: Do not filter detections based on overlap.
+        NON_MAX_SUPPRESSION: Filter detections using non-max suppression. This means,
+            detections that overlap by more than a set threshold will be discarded,
+            except for the one with the highest confidence.
+        NON_MAX_MERGE: Merge detections with non-max merging. This means,
+            detections that overlap by more than a set threshold will be merged
+            into a single detection.
+    """
+
+    NONE = "none"
+    NON_MAX_SUPPRESSION = "non_max_suppression"
+    NON_MAX_MERGE = "non_max_merge"
+
+
+def validate_overlap_filter(
+    strategy: Union[OverlapFilter, str],
+) -> OverlapFilter:
+    if isinstance(strategy, str):
+        try:
+            strategy = OverlapFilter(strategy.lower())
+        except ValueError:
+            raise ValueError(
+                f"Invalid strategy value: {strategy}. Must be one of "
+                f"{[e.value for e in OverlapFilter]}"
+            )
+    return strategy
@@ -1,23 +1,42 @@
+import warnings
 from concurrent.futures import ThreadPoolExecutor, as_completed
-from typing import Callable, Optional, Tuple
+from typing import Callable, Optional, Tuple, Union

 import numpy as np

 from supervision.detection.core import Detections
-from supervision.detection.utils import move_boxes
+from supervision.detection.overlap_filter import OverlapFilter, validate_overlap_filter
+from supervision.detection.utils import move_boxes, move_masks
 from supervision.utils.image import crop_image
+from supervision.utils.internal import SupervisionWarnings


-def move_detections(detections: Detections, offset: np.array) -> Detections:
+def move_detections(
+    detections: Detections,
+    offset: np.ndarray,
+    resolution_wh: Optional[Tuple[int, int]] = None,
+) -> Detections:
    """
    Args:
        detections (sv.Detections): Detections object to be moved.
-        offset (np.array): An array of shape `(2,)` containing offset values in format
+        offset (np.ndarray): An array of shape `(2,)` containing offset values in format
            is `[dx, dy]`.
+        resolution_wh (Tuple[int, int]): The width and height of the desired mask
+            resolution. Required for segmentation detections.
+
    Returns:
        (sv.Detections) repositioned Detections object.
    """
    detections.xyxy = move_boxes(xyxy=detections.xyxy, offset=offset)
+    if detections.mask is not None:
+        if resolution_wh is None:
+            raise ValueError(
+                "Resolution width and height are required for moving segmentation "
+                "detections. This should be the same as (width, height) of image shape."
+            )
+        detections.mask = move_masks(
+            masks=detections.mask, offset=offset, resolution_wh=resolution_wh
+        )
    return detections


@@ -34,8 +53,10 @@ class InferenceSlicer:
            `(width, height)`.
        overlap_ratio_wh (Tuple[float, float]): Overlap ratio between consecutive
            slices in the format `(width_ratio, height_ratio)`.
-        iou_threshold (Optional[float]): Intersection over Union (IoU) threshold
-            used for non-max suppression.
+        overlap_filter_strategy (Union[OverlapFilter, str]): Strategy for
+            filtering or merging overlapping detections in slices.
+        iou_threshold (float): Intersection over Union (IoU) threshold
+            used when filtering by overlap.
        callback (Callable): A function that performs inference on a given image
            slice and returns detections.
        thread_workers (int): Number of threads for parallel execution.
@@ -52,12 +73,18 @@ class InferenceSlicer:
        callback: Callable[[np.ndarray], Detections],
        slice_wh: Tuple[int, int] = (320, 320),
        overlap_ratio_wh: Tuple[float, float] = (0.2, 0.2),
-        iou_threshold: Optional[float] = 0.5,
+        overlap_filter_strategy: Union[
+            OverlapFilter, str
+        ] = OverlapFilter.NON_MAX_SUPPRESSION,
+        iou_threshold: float = 0.5,
        thread_workers: int = 1,
    ):
+        overlap_filter_strategy = validate_overlap_filter(overlap_filter_strategy)
+
        self.slice_wh = slice_wh
        self.overlap_ratio_wh = overlap_ratio_wh
        self.iou_threshold = iou_threshold
+        self.overlap_filter_strategy = overlap_filter_strategy
        self.callback = callback
        self.thread_workers = thread_workers

@@ -88,7 +115,10 @@ class InferenceSlicer:
                result = model(image_slice)[0]
                return sv.Detections.from_ultralytics(result)

-            slicer = sv.InferenceSlicer(callback = callback)
+            slicer = sv.InferenceSlicer(
+                callback=callback,
+                overlap_filter_strategy=sv.OverlapFilter.NON_MAX_SUPPRESSION,
+            )

            detections = slicer(image)
            ```
@@ -108,9 +138,19 @@ class InferenceSlicer:
            for future in as_completed(futures):
                detections_list.append(future.result())

-        return Detections.merge(detections_list=detections_list).with_nms(
-            threshold=self.iou_threshold
-        )
+        merged = Detections.merge(detections_list=detections_list)
+        if self.overlap_filter_strategy == OverlapFilter.NONE:
+            return merged
+        elif self.overlap_filter_strategy == OverlapFilter.NON_MAX_SUPPRESSION:
+            return merged.with_nms(threshold=self.iou_threshold)
+        elif self.overlap_filter_strategy == OverlapFilter.NON_MAX_MERGE:
+            return merged.with_nmm(threshold=self.iou_threshold)
+        else:
+            warnings.warn(
+                f"Invalid overlap filter strategy: {self.overlap_filter_strategy}",
+                category=SupervisionWarnings,
+            )
+            return merged

    def _run_callback(self, image, offset) -> Detections:
        """
@@ -126,7 +166,10 @@ class InferenceSlicer:
        """
        image_slice = crop_image(image=image, xyxy=offset)
        detections = self.callback(image_slice)
-        detections = move_detections(detections=detections, offset=offset[:2])
+        resolution_wh = (image.shape[1], image.shape[0])
+        detections = move_detections(
+            detections=detections, offset=offset[:2], resolution_wh=resolution_wh
+        )

        return detections

@@ -1,8 +1,10 @@
+import warnings
 from dataclasses import replace
 from typing import Iterable, Optional, Tuple

 import cv2
 import numpy as np
+import numpy.typing as npt

 from supervision import Detections
 from supervision.detection.utils import clip_boxes, polygon_to_mask
@@ -10,7 +12,7 @@ from supervision.draw.color import Color
 from supervision.draw.utils import draw_polygon, draw_text
 from supervision.geometry.core import Position
 from supervision.geometry.utils import get_polygon_center
-from supervision.utils.internal import deprecated_parameter
+from supervision.utils.internal import SupervisionWarnings, deprecated_parameter


 class PolygonZone:
@@ -20,7 +22,6 @@ class PolygonZone:
    Attributes:
        polygon (np.ndarray): A polygon represented by a numpy array of shape
            `(N, 2)`, containing the `x`, `y` coordinates of the points.
-        frame_resolution_wh (Tuple[int, int]): The frame resolution (width, height)
        triggering_anchors (Iterable[sv.Position]): A list of positions specifying
            which anchors of the detections bounding box to consider when deciding on
            whether the detection fits within the PolygonZone
@@ -39,22 +40,32 @@ class PolygonZone:
    )
    def __init__(
        self,
-        polygon: np.ndarray,
-        frame_resolution_wh: Tuple[int, int],
+        polygon: npt.NDArray[np.int64],
+        frame_resolution_wh: Optional[Tuple[int, int]] = None,
        triggering_anchors: Iterable[Position] = (Position.BOTTOM_CENTER,),
    ):
+        if frame_resolution_wh is not None:
+            warnings.warn(
+                "The `frame_resolution_wh` parameter is no longer required and will be "
+                "dropped in version supervision-0.24.0. The mask resolution is now "
+                "calculated automatically based on the polygon coordinates.",
+                category=SupervisionWarnings,
+            )
+
        self.polygon = polygon.astype(int)
-        self.frame_resolution_wh = frame_resolution_wh
        self.triggering_anchors = triggering_anchors
+        if not list(self.triggering_anchors):
+            raise ValueError("Triggering anchors cannot be empty.")

        self.current_count = 0

-        width, height = frame_resolution_wh
+        x_max, y_max = np.max(polygon, axis=0)
+        self.frame_resolution_wh = (x_max + 1, y_max + 1)
        self.mask = polygon_to_mask(
-            polygon=polygon, resolution_wh=(width + 1, height + 1)
+            polygon=polygon, resolution_wh=(x_max + 2, y_max + 2)
        )

-    def trigger(self, detections: Detections) -> np.ndarray:
+    def trigger(self, detections: Detections) -> npt.NDArray[np.bool_]:
        """
        Determines if the detections are within the polygon zone.

@@ -78,13 +89,13 @@ class PolygonZone:
            ]
        )

-        is_in_zone = (
+        is_in_zone: npt.NDArray[np.bool_] = (
            self.mask[all_clipped_anchors[:, :, 1], all_clipped_anchors[:, :, 0]]
            .transpose()
            .astype(bool)
        )
-        is_in_zone = np.all(is_in_zone, axis=1)

+        is_in_zone: npt.NDArray[np.bool_] = np.all(is_in_zone, axis=1)
        self.current_count = int(np.sum(is_in_zone))
        return is_in_zone.astype(bool)

@@ -1,3 +1,4 @@
+import warnings
 from collections import defaultdict, deque
 from copy import deepcopy
 from typing import Optional
@@ -5,6 +6,7 @@ from typing import Optional
 import numpy as np

 from supervision.detection.core import Detections
+from supervision.utils.internal import SupervisionWarnings


 class DetectionsSmoother:
@@ -70,16 +72,16 @@ class DetectionsSmoother:
        """

        if detections.tracker_id is None:
-            print(
+            warnings.warn(
                "Smoothing skipped. DetectionsSmoother requires tracker_id. Refer to "
-                "https://supervision.roboflow.com/latest/trackers for more information."
+                "https://supervision.roboflow.com/latest/trackers for more "
+                "information.",
+                category=SupervisionWarnings,
            )
            return detections

        for detection_idx in range(len(detections)):
            tracker_id = detections.tracker_id[detection_idx]
-            if tracker_id is None:
-                continue

            self.tracks[tracker_id].append(detections[detection_idx])

@@ -1,10 +1,12 @@
 from itertools import chain
-from typing import Any, Dict, List, Optional, Tuple, Union
+from typing import Dict, List, Optional, Tuple, Union

 import cv2
 import numpy as np
+import numpy.typing as npt

 from supervision.config import CLASS_NAME_DATA_FIELD
+from supervision.geometry.core import Vector

 MIN_POLYGON_POINT_COUNT = 3

@@ -56,7 +58,9 @@ def box_iou_batch(boxes_true: np.ndarray, boxes_detection: np.ndarray) -> np.nda
    bottom_right = np.minimum(boxes_true[:, None, 2:], boxes_detection[:, 2:])

    area_inter = np.prod(np.clip(bottom_right - top_left, a_min=0, a_max=None), 2)
-    return area_inter / (area_true[:, None] + area_detection - area_inter)
+    ious = area_inter / (area_true[:, None] + area_detection - area_inter)
+    ious = np.nan_to_num(ious)
+    return ious


 def _mask_iou_batch_split(
@@ -136,144 +140,6 @@ def mask_iou_batch(
    return np.vstack(ious)


-def resize_masks(masks: np.ndarray, max_dimension: int = 640) -> np.ndarray:
-    """
-    Resize all masks in the array to have a maximum dimension of max_dimension,
-    maintaining aspect ratio.
-
-    Args:
-        masks (np.ndarray): 3D array of binary masks with shape (N, H, W).
-        max_dimension (int): The maximum dimension for the resized masks.
-
-    Returns:
-        np.ndarray: Array of resized masks.
-    """
-    max_height = np.max(masks.shape[1])
-    max_width = np.max(masks.shape[2])
-    scale = min(max_dimension / max_height, max_dimension / max_width)
-
-    new_height = int(scale * max_height)
-    new_width = int(scale * max_width)
-
-    x = np.linspace(0, max_width - 1, new_width).astype(int)
-    y = np.linspace(0, max_height - 1, new_height).astype(int)
-    xv, yv = np.meshgrid(x, y)
-
-    resized_masks = masks[:, yv, xv]
-
-    resized_masks = resized_masks.reshape(masks.shape[0], new_height, new_width)
-    return resized_masks
-
-
-def mask_non_max_suppression(
-    predictions: np.ndarray,
-    masks: np.ndarray,
-    iou_threshold: float = 0.5,
-    mask_dimension: int = 640,
-) -> np.ndarray:
-    """
-    Perform Non-Maximum Suppression (NMS) on segmentation predictions.
-
-    Args:
-        predictions (np.ndarray): A 2D array of object detection predictions in
-            the format of `(x_min, y_min, x_max, y_max, score)`
-            or `(x_min, y_min, x_max, y_max, score, class)`. Shape: `(N, 5)` or
-            `(N, 6)`, where N is the number of predictions.
-        masks (np.ndarray): A 3D array of binary masks corresponding to the predictions.
-            Shape: `(N, H, W)`, where N is the number of predictions, and H, W are the
-            dimensions of each mask.
-        iou_threshold (float, optional): The intersection-over-union threshold
-            to use for non-maximum suppression.
-        mask_dimension (int, optional): The dimension to which the masks should be
-            resized before computing IOU values. Defaults to 640.
-
-    Returns:
-        np.ndarray: A boolean array indicating which predictions to keep after
-            non-maximum suppression.
-
-    Raises:
-        AssertionError: If `iou_threshold` is not within the closed
-        range from `0` to `1`.
-    """
-    assert 0 <= iou_threshold <= 1, (
-        "Value of `iou_threshold` must be in the closed range from 0 to 1, "
-        f"{iou_threshold} given."
-    )
-    rows, columns = predictions.shape
-
-    if columns == 5:
-        predictions = np.c_[predictions, np.zeros(rows)]
-
-    sort_index = predictions[:, 4].argsort()[::-1]
-    predictions = predictions[sort_index]
-    masks = masks[sort_index]
-    masks_resized = resize_masks(masks, mask_dimension)
-    ious = mask_iou_batch(masks_resized, masks_resized)
-    categories = predictions[:, 5]
-
-    keep = np.ones(rows, dtype=bool)
-    for i in range(rows):
-        if keep[i]:
-            condition = (ious[i] > iou_threshold) & (categories[i] == categories)
-            keep[i + 1 :] = np.where(condition[i + 1 :], False, keep[i + 1 :])
-
-    return keep[sort_index.argsort()]
-
-
-def box_non_max_suppression(
-    predictions: np.ndarray, iou_threshold: float = 0.5
-) -> np.ndarray:
-    """
-    Perform Non-Maximum Suppression (NMS) on object detection predictions.
-
-    Args:
-        predictions (np.ndarray): An array of object detection predictions in
-            the format of `(x_min, y_min, x_max, y_max, score)`
-            or `(x_min, y_min, x_max, y_max, score, class)`.
-        iou_threshold (float, optional): The intersection-over-union threshold
-            to use for non-maximum suppression.
-
-    Returns:
-        np.ndarray: A boolean array indicating which predictions to keep after n
-            on-maximum suppression.
-
-    Raises:
-        AssertionError: If `iou_threshold` is not within the
-            closed range from `0` to `1`.
-    """
-    assert 0 <= iou_threshold <= 1, (
-        "Value of `iou_threshold` must be in the closed range from 0 to 1, "
-        f"{iou_threshold} given."
-    )
-    rows, columns = predictions.shape
-
-    # add column #5 - category filled with zeros for agnostic nms
-    if columns == 5:
-        predictions = np.c_[predictions, np.zeros(rows)]
-
-    # sort predictions column #4 - score
-    sort_index = np.flip(predictions[:, 4].argsort())
-    predictions = predictions[sort_index]
-
-    boxes = predictions[:, :4]
-    categories = predictions[:, 5]
-    ious = box_iou_batch(boxes, boxes)
-    ious = ious - np.eye(rows)
-
-    keep = np.ones(rows, dtype=bool)
-
-    for index, (iou, category) in enumerate(zip(ious, categories)):
-        if not keep[index]:
-            continue
-
-        # drop detections with iou > iou_threshold and
-        # same category as current detections
-        condition = (iou > iou_threshold) & (categories == category)
-        keep = keep & ~condition
-
-    return keep[sort_index.argsort()]
-
-
 def clip_boxes(xyxy: np.ndarray, resolution_wh: Tuple[int, int]) -> np.ndarray:
    """
    Clips bounding boxes coordinates to fit within the frame resolution.
@@ -289,6 +155,25 @@ def clip_boxes(xyxy: np.ndarray, resolution_wh: Tuple[int, int]) -> np.ndarray:
        np.ndarray: A numpy array of shape `(N, 4)` where each row
            corresponds to a bounding box with coordinates clipped to fit
            within the frame resolution.
+
+    Examples:
+        ```python
+        import numpy as np
+        import supervision as sv
+
+        xyxy = np.array([
+            [10, 20, 300, 200],
+            [15, 25, 350, 450],
+            [-10, -20, 30, 40]
+        ])
+
+        sv.clip_boxes(xyxy=xyxy, resolution_wh=(320, 240))
+        # array([
+        #     [ 10,  20, 300, 200],
+        #     [ 15,  25, 320, 240],
+        #     [  0,   0,  30,  40]
+        # ])
+        ```
    """
    result = np.copy(xyxy)
    width, height = resolution_wh
@@ -297,6 +182,52 @@ def clip_boxes(xyxy: np.ndarray, resolution_wh: Tuple[int, int]) -> np.ndarray:
    return result


+def pad_boxes(xyxy: np.ndarray, px: int, py: Optional[int] = None) -> np.ndarray:
+    """
+    Pads bounding boxes coordinates with a constant padding.
+
+    Args:
+        xyxy (np.ndarray): A numpy array of shape `(N, 4)` where each
+            row corresponds to a bounding box in the format
+            `(x_min, y_min, x_max, y_max)`.
+        px (int): The padding value to be added to both the left and right sides of
+            each bounding box.
+        py (Optional[int]): The padding value to be added to both the top and bottom
+            sides of each bounding box. If not provided, `px` will be used for both
+            dimensions.
+
+    Returns:
+        np.ndarray: A numpy array of shape `(N, 4)` where each row corresponds to a
+            bounding box with coordinates padded according to the provided padding
+            values.
+
+    Examples:
+        ```python
+        import numpy as np
+        import supervision as sv
+
+        xyxy = np.array([
+            [10, 20, 30, 40],
+            [15, 25, 35, 45]
+        ])
+
+        sv.pad_boxes(xyxy=xyxy, px=5, py=10)
+        # array([
+        #     [ 5, 10, 35, 50],
+        #     [10, 15, 40, 55]
+        # ])
+        ```
+    """
+    if py is None:
+        py = px
+
+    result = xyxy.copy()
+    result[:, [0, 1]] -= [px, py]
+    result[:, [2, 3]] += [px, py]
+
+    return result
+
+
 def xywh_to_xyxy(boxes_xywh: np.ndarray) -> np.ndarray:
    xyxy = boxes_xywh.copy()
    xyxy[:, 2] = boxes_xywh[:, 0] + boxes_xywh[:, 2]
@@ -317,7 +248,7 @@ def mask_to_xyxy(masks: np.ndarray) -> np.ndarray:
            `(x_min, y_min, x_max, y_max)` for each mask
    """
    n = masks.shape[0]
-    bboxes = np.zeros((n, 4), dtype=int)
+    xyxy = np.zeros((n, 4), dtype=int)

    for i, mask in enumerate(masks):
        rows, cols = np.where(mask)
@@ -325,9 +256,9 @@ def mask_to_xyxy(masks: np.ndarray) -> np.ndarray:
        if len(rows) > 0 and len(cols) > 0:
            x_min, x_max = np.min(cols), np.max(cols)
            y_min, y_max = np.min(rows), np.max(rows)
-            bboxes[i, :] = [x_min, y_min, x_max, y_max]
+            xyxy[i, :] = [x_min, y_min, x_max, y_max]

-    return bboxes
+    return xyxy


 def mask_to_polygons(mask: np.ndarray) -> List[np.ndarray]:
@@ -500,7 +431,7 @@ def process_roboflow_result(
    np.ndarray,
    Optional[np.ndarray],
    Optional[np.ndarray],
-    Dict[str, List[np.ndarray]],
+    Dict[str, Union[List[np.ndarray], np.ndarray]],
 ]:
    if not roboflow_result["predictions"]:
        return (
@@ -563,59 +494,103 @@ def process_roboflow_result(
    return xyxy, confidence, class_id, masks, tracker_id, data


-def move_boxes(xyxy: np.ndarray, offset: np.ndarray) -> np.ndarray:
+def move_boxes(
+    xyxy: npt.NDArray[np.float64], offset: npt.NDArray[np.int32]
+) -> npt.NDArray[np.float64]:
    """
    Parameters:
-        xyxy (np.ndarray): An array of shape `(n, 4)` containing the bounding boxes
-            coordinates in format `[x1, y1, x2, y2]`
+        xyxy (npt.NDArray[np.float64]): An array of shape `(n, 4)` containing the
+            bounding boxes coordinates in format `[x1, y1, x2, y2]`
        offset (np.array): An array of shape `(2,)` containing offset values in format
            is `[dx, dy]`.

    Returns:
-        np.ndarray: Repositioned bounding boxes.
+        npt.NDArray[np.float64]: Repositioned bounding boxes.

-    Example:
+    Examples:
        ```python
        import numpy as np
        import supervision as sv

-        boxes = np.array([[10, 10, 20, 20], [30, 30, 40, 40]])
+        xyxy = np.array([
+            [10, 10, 20, 20],
+            [30, 30, 40, 40]
+        ])
        offset = np.array([5, 5])
-        moved_box = sv.move_boxes(boxes, offset)
-        print(moved_box)
-        # np.array([
+
+        sv.move_boxes(xyxy=xyxy, offset=offset)
+        # array([
        #    [15, 15, 25, 25],
-        #     [35, 35, 45, 45]
+        #    [35, 35, 45, 45]
        # ])
        ```
    """
    return xyxy + np.hstack([offset, offset])


-def scale_boxes(xyxy: np.ndarray, factor: float) -> np.ndarray:
+def move_masks(
+    masks: npt.NDArray[np.bool_],
+    offset: npt.NDArray[np.int32],
+    resolution_wh: Tuple[int, int],
+) -> npt.NDArray[np.bool_]:
+    """
+    Offset the masks in an array by the specified (x, y) amount.
+
+    Args:
+        masks (npt.NDArray[np.bool_]): A 3D array of binary masks corresponding to the
+            predictions. Shape: `(N, H, W)`, where N is the number of predictions, and
+            H, W are the dimensions of each mask.
+        offset (npt.NDArray[np.int32]): An array of shape `(2,)` containing non-negative
+            int values `[dx, dy]`.
+        resolution_wh (Tuple[int, int]): The width and height of the desired mask
+            resolution.
+
+    Returns:
+        (npt.NDArray[np.bool_]) repositioned masks, optionally padded to the specified
+            shape.
+    """
+
+    if offset[0] < 0 or offset[1] < 0:
+        raise ValueError(f"Offset values must be non-negative integers. Got: {offset}")
+
+    mask_array = np.full((masks.shape[0], resolution_wh[1], resolution_wh[0]), False)
+    mask_array[
+        :,
+        offset[1] : masks.shape[1] + offset[1],
+        offset[0] : masks.shape[2] + offset[0],
+    ] = masks
+
+    return mask_array
+
+
+def scale_boxes(
+    xyxy: npt.NDArray[np.float64], factor: float
+) -> npt.NDArray[np.float64]:
    """
    Scale the dimensions of bounding boxes.

    Parameters:
-        xyxy (np.ndarray): An array of shape `(n, 4)` containing the bounding boxes
-            coordinates in format `[x1, y1, x2, y2]`
+        xyxy (npt.NDArray[np.float64]): An array of shape `(n, 4)` containing the
+            bounding boxes coordinates in format `[x1, y1, x2, y2]`
        factor (float): A float value representing the factor by which the box
            dimensions are scaled. A factor greater than 1 enlarges the boxes, while a
            factor less than 1 shrinks them.

    Returns:
-        np.ndarray: Scaled bounding boxes.
+        npt.NDArray[np.float64]: Scaled bounding boxes.

-    Example:
+    Examples:
        ```python
        import numpy as np
        import supervision as sv

-        boxes = np.array([[10, 10, 20, 20], [30, 30, 40, 40]])
-        factor = 1.5
-        scaled_bb = sv.scale_boxes(boxes, factor)
-        print(scaled_bb)
-        # np.array([
+        xyxy = np.array([
+            [10, 10, 20, 20],
+            [30, 30, 40, 40]
+        ])
+
+        sv.scale_boxes(xyxy=xyxy, factor=1.5)
+        # array([
        #    [ 7.5,  7.5, 22.5, 22.5],
        #    [27.5, 27.5, 42.5, 42.5]
        # ])
@@ -656,102 +631,6 @@ def calculate_masks_centroids(masks: np.ndarray) -> np.ndarray:
    return np.column_stack((centroid_x, centroid_y)).astype(int)


-def validate_xyxy(xyxy: Any) -> None:
-    expected_shape = "(_, 4)"
-    actual_shape = str(getattr(xyxy, "shape", None))
-    is_valid = isinstance(xyxy, np.ndarray) and xyxy.ndim == 2 and xyxy.shape[1] == 4
-    if not is_valid:
-        raise ValueError(
-            f"xyxy must be a 2D np.ndarray with shape {expected_shape}, but got shape "
-            f"{actual_shape}"
-        )
-
-
-def validate_mask(mask: Any, n: int) -> None:
-    expected_shape = f"({n}, H, W)"
-    actual_shape = str(getattr(mask, "shape", None))
-    is_valid = mask is None or (
-        isinstance(mask, np.ndarray) and len(mask.shape) == 3 and mask.shape[0] == n
-    )
-    if not is_valid:
-        raise ValueError(
-            f"mask must be a 3D np.ndarray with shape {expected_shape}, but got shape "
-            f"{actual_shape}"
-        )
-
-
-def validate_class_id(class_id: Any, n: int) -> None:
-    expected_shape = f"({n},)"
-    actual_shape = str(getattr(class_id, "shape", None))
-    is_valid = class_id is None or (
-        isinstance(class_id, np.ndarray) and class_id.shape == (n,)
-    )
-    if not is_valid:
-        raise ValueError(
-            f"class_id must be a 1D np.ndarray with shape {expected_shape}, but got "
-            f"shape {actual_shape}"
-        )
-
-
-def validate_confidence(confidence: Any, n: int) -> None:
-    expected_shape = f"({n},)"
-    actual_shape = str(getattr(confidence, "shape", None))
-    is_valid = confidence is None or (
-        isinstance(confidence, np.ndarray) and confidence.shape == (n,)
-    )
-    if not is_valid:
-        raise ValueError(
-            f"confidence must be a 1D np.ndarray with shape {expected_shape}, but got "
-            f"shape {actual_shape}"
-        )
-
-
-def validate_tracker_id(tracker_id: Any, n: int) -> None:
-    expected_shape = f"({n},)"
-    actual_shape = str(getattr(tracker_id, "shape", None))
-    is_valid = tracker_id is None or (
-        isinstance(tracker_id, np.ndarray) and tracker_id.shape == (n,)
-    )
-    if not is_valid:
-        raise ValueError(
-            f"tracker_id must be a 1D np.ndarray with shape {expected_shape}, but got "
-            f"shape {actual_shape}"
-        )
-
-
-def validate_data(data: Dict[str, Any], n: int) -> None:
-    for key, value in data.items():
-        if isinstance(value, list):
-            if len(value) != n:
-                raise ValueError(f"Length of list for key '{key}' must be {n}")
-        elif isinstance(value, np.ndarray):
-            if value.ndim == 1 and value.shape[0] != n:
-                raise ValueError(f"Shape of np.ndarray for key '{key}' must be ({n},)")
-            elif value.ndim > 1 and value.shape[0] != n:
-                raise ValueError(
-                    f"First dimension of np.ndarray for key '{key}' must have size {n}"
-                )
-        else:
-            raise ValueError(f"Value for key '{key}' must be a list or np.ndarray")
-
-
-def validate_detections_fields(
-    xyxy: Any,
-    mask: Any,
-    class_id: Any,
-    confidence: Any,
-    tracker_id: Any,
-    data: Dict[str, Any],
-) -> None:
-    validate_xyxy(xyxy)
-    n = len(xyxy)
-    validate_mask(mask, n)
-    validate_class_id(class_id, n)
-    validate_confidence(confidence, n)
-    validate_tracker_id(tracker_id, n)
-    validate_data(data, n)
-
-
 def is_data_equal(data_a: Dict[str, np.ndarray], data_b: Dict[str, np.ndarray]) -> bool:
    """
    Compares the data payloads of two Detections instances.
@@ -768,17 +647,19 @@ def is_data_equal(data_a: Dict[str, np.ndarray], data_b: Dict[str, np.ndarray])


 def merge_data(
-    data_list: List[Dict[str, Union[np.ndarray, List]]],
-) -> Dict[str, Union[np.ndarray, List]]:
+    data_list: List[Dict[str, Union[npt.NDArray[np.generic], List]]],
+) -> Dict[str, Union[npt.NDArray[np.generic], List]]:
    """
    Merges the data payloads of a list of Detections instances.

    Args:
-        data_list: The data payloads of the instances.
+        data_list: The data payloads of the Detections instances. Each data payload
+            is a dictionary with the same keys, and the values are either lists or
+            npt.NDArray[np.generic].

    Returns:
        A single data payload containing the merged data, preserving the original data
-            types (list or np.ndarray).
+            types (list or npt.NDArray[np.generic]).

    Raises:
        ValueError: If data values within a single object have different lengths or if
@@ -799,9 +680,8 @@ def merge_data(
            )

    merged_data = {key: [] for key in all_keys_sets[0]}
-
    for data in data_list:
-        for key in merged_data:
+        for key in data:
            merged_data[key].append(data[key])

    for key in merged_data:
@@ -845,8 +725,15 @@ def get_data_item(
        elif isinstance(value, list):
            if isinstance(index, slice):
                subset_data[key] = value[index]
-            elif isinstance(index, (list, np.ndarray)):
+            elif isinstance(index, list):
                subset_data[key] = [value[i] for i in index]
+            elif isinstance(index, np.ndarray):
+                if index.dtype == bool:
+                    subset_data[key] = [
+                        value[i] for i, index_value in enumerate(index) if index_value
+                    ]
+                else:
+                    subset_data[key] = [value[i] for i in index]
            elif isinstance(index, int):
                subset_data[key] = [value[index]]
            else:
@@ -855,3 +742,138 @@ def get_data_item(
            raise TypeError(f"Unsupported data type for key '{key}': {type(value)}")

    return subset_data
+
+
+def contains_holes(mask: npt.NDArray[np.bool_]) -> bool:
+    """
+    Checks if the binary mask contains holes (background pixels fully enclosed by
+    foreground pixels).
+
+    Args:
+        mask (npt.NDArray[np.bool_]): 2D binary mask where `True` indicates foreground
+            object and `False` indicates background.
+
+    Returns:
+        True if holes are detected, False otherwise.
+
+    Examples:
+        ```python
+        import numpy as np
+        import supervision as sv
+
+        mask = np.array([
+            [0, 0, 0, 0, 0],
+            [0, 1, 1, 1, 0],
+            [0, 1, 0, 1, 0],
+            [0, 1, 1, 1, 0],
+            [0, 0, 0, 0, 0]
+        ]).astype(bool)
+
+        sv.contains_holes(mask=mask)
+        # True
+
+        mask = np.array([
+            [0, 0, 0, 0, 0],
+            [0, 1, 1, 1, 0],
+            [0, 1, 1, 1, 0],
+            [0, 1, 1, 1, 0],
+            [0, 0, 0, 0, 0]
+        ]).astype(bool)
+
+        sv.contains_holes(mask=mask)
+        # False
+        ```
+
+    ![contains_holes](https://media.roboflow.com/supervision-docs/contains-holes.png){ align=center width="800" }
+    """  # noqa E501 // docs
+    mask_uint8 = mask.astype(np.uint8)
+    _, hierarchy = cv2.findContours(mask_uint8, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
+
+    if hierarchy is not None:
+        parent_contour_index = 3
+        for h in hierarchy[0]:
+            if h[parent_contour_index] != -1:
+                return True
+    return False
+
+
+def contains_multiple_segments(
+    mask: npt.NDArray[np.bool_], connectivity: int = 4
+) -> bool:
+    """
+    Checks if the binary mask contains multiple unconnected foreground segments.
+
+    Args:
+        mask (npt.NDArray[np.bool_]): 2D binary mask where `True` indicates foreground
+            object and `False` indicates background.
+        connectivity (int) : Default: 4 is 4-way connectivity, which means that
+            foreground pixels are the part of the same segment/component
+            if their edges touch.
+            Alternatively: 8 for 8-way connectivity, when foreground pixels are
+            connected by their edges or corners touch.
+
+    Returns:
+        True when the mask contains multiple not connected components, False otherwise.
+
+    Raises:
+        ValueError: If connectivity(int) parameter value is not 4 or 8.
+
+    Examples:
+        ```python
+        import numpy as np
+        import supervision as sv
+
+        mask = np.array([
+            [0, 0, 0, 0, 0, 0],
+            [0, 1, 1, 0, 1, 1],
+            [0, 1, 1, 0, 1, 1],
+            [0, 0, 0, 0, 0, 0],
+            [0, 1, 1, 1, 0, 0],
+            [0, 1, 1, 1, 0, 0]
+        ]).astype(bool)
+
+        sv.contains_multiple_segments(mask=mask, connectivity=4)
+        # True
+
+        mask = np.array([
+            [0, 0, 0, 0, 0, 0],
+            [0, 1, 1, 1, 1, 1],
+            [0, 1, 1, 1, 1, 1],
+            [0, 1, 1, 1, 1, 1],
+            [0, 1, 1, 1, 1, 1],
+            [0, 0, 0, 0, 0, 0]
+        ]).astype(bool)
+
+        sv.contains_multiple_segments(mask=mask, connectivity=4)
+        # False
+        ```
+
+    ![contains_multiple_segments](https://media.roboflow.com/supervision-docs/contains-multiple-segments.png){ align=center width="800" }
+    """  # noqa E501 // docs
+    if connectivity != 4 and connectivity != 8:
+        raise ValueError(
+            "Incorrect connectivity value. Possible connectivity values: 4 or 8."
+        )
+    mask_uint8 = mask.astype(np.uint8)
+    labels = np.zeros_like(mask_uint8, dtype=np.int32)
+    number_of_labels, _ = cv2.connectedComponents(
+        mask_uint8, labels, connectivity=connectivity
+    )
+    return number_of_labels > 2
+
+
+def cross_product(anchors: np.ndarray, vector: Vector) -> np.ndarray:
+    """
+    Get array of cross products of each anchor with a vector.
+    Args:
+        anchors: Array of anchors of shape (number of anchors, detections, 2)
+        vector: Vector to calculate cross product with
+
+    Returns:
+        Array of cross products of shape (number of anchors, detections)
+    """
+    vector_at_zero = np.array(
+        [vector.end.x - vector.start.x, vector.end.y - vector.start.y]
+    )
+    vector_start = np.array([vector.start.x, vector.start.y])
+    return np.cross(vector_at_zero, anchors - vector_start)
@@ -1,7 +1,7 @@
 from __future__ import annotations

 from dataclasses import dataclass
-from typing import List, Tuple
+from typing import List, Tuple, Union

 import matplotlib.pyplot as plt

@@ -104,10 +104,13 @@ class Color:
        Create a Color instance from a hex string.

        Args:
-            color_hex (str): Hex string of the color.
+            color_hex (str): The hex string representing the color. This string can
+                start with '#' followed by either 3 or 6 hexadecimal characters. In
+                case of 3 characters, each character is repeated to form the full
+                6-character hex code.

        Returns:
-            Color: Instance representing the color.
+            Color: An instance representing the color.

        Example:
            ```python
@@ -115,6 +118,9 @@ class Color:

            sv.Color.from_hex('#ff00ff')
            # Color(r=255, g=0, b=255)
+
+            sv.Color.from_hex('#f0f')
+            # Color(r=255, g=0, b=255)
            ```
        """
        _validate_color_hex(color_hex)
@@ -124,6 +130,52 @@ class Color:
        r, g, b = (int(color_hex[i : i + 2], 16) for i in range(0, 6, 2))
        return cls(r, g, b)

+    @classmethod
+    def from_rgb_tuple(cls, color_tuple: Tuple[int, int, int]) -> Color:
+        """
+        Create a Color instance from an RGB tuple.
+
+        Args:
+            color_tuple (Tuple[int, int, int]): A tuple representing the color in RGB
+                format, where each element is an integer in the range 0-255.
+
+        Returns:
+            Color: An instance representing the color.
+
+        Example:
+            ```python
+            import supervision as sv
+
+            sv.Color.from_rgb_tuple((255, 255, 0))
+            # Color(r=255, g=255, b=0)
+            ```
+        """
+        r, g, b = color_tuple
+        return cls(r=r, g=g, b=b)
+
+    @classmethod
+    def from_bgr_tuple(cls, color_tuple: Tuple[int, int, int]) -> Color:
+        """
+        Create a Color instance from a BGR tuple.
+
+        Args:
+            color_tuple (Tuple[int, int, int]): A tuple representing the color in BGR
+                format, where each element is an integer in the range 0-255.
+
+        Returns:
+            Color: An instance representing the color.
+
+        Example:
+            ```python
+            import supervision as sv
+
+            sv.Color.from_bgr_tuple((0, 255, 255))
+            # Color(r=255, g=255, b=0)
+            ```
+        """
+        b, g, r = color_tuple
+        return cls(r=r, g=g, b=b)
+
    def as_hex(self) -> str:
        """
        Converts the Color instance to a hex string.
@@ -176,31 +228,31 @@ class Color:
        return self.b, self.g, self.r

    @classproperty
-    def WHITE(cls):
+    def WHITE(cls) -> Color:
        return Color.from_hex("#FFFFFF")

    @classproperty
-    def BLACK(cls):
+    def BLACK(cls) -> Color:
        return Color.from_hex("#000000")

    @classproperty
-    def RED(cls):
+    def RED(cls) -> Color:
        return Color.from_hex("#FF0000")

    @classproperty
-    def GREEN(cls):
+    def GREEN(cls) -> Color:
        return Color.from_hex("#00FF00")

    @classproperty
-    def BLUE(cls):
+    def BLUE(cls) -> Color:
        return Color.from_hex("#0000FF")

    @classproperty
-    def YELLOW(cls):
+    def YELLOW(cls) -> Color:
        return Color.from_hex("#FFFF00")

    @classproperty
-    def ROBOFLOW(cls):
+    def ROBOFLOW(cls) -> Color:
        return Color.from_hex("#A351FB")

    @classmethod
@@ -396,3 +448,19 @@ class ColorPalette:
            raise ValueError("idx argument should not be negative")
        idx = idx % len(self.colors)
        return self.colors[idx]
+
+
+def unify_to_bgr(color: Union[Tuple[int, int, int], Color]) -> Tuple[int, int, int]:
+    """
+    Converts a color input in multiple formats to a standardized BGR format.
+
+    Args:
+        color (Union[Tuple[int, int, int], Color]): The color input to be converted,
+            which can be either a tuple of RGB values or an instance of a Color class.
+
+    Returns:
+        Tuple[int, int, int]: The color in BGR format as a tuple of three integers.
+    """
+    if issubclass(type(color), Color):
+        return color.as_bgr()
+    return color
@@ -81,6 +81,58 @@ def draw_filled_rectangle(scene: np.ndarray, rect: Rect, color: Color) -> np.nda
    return scene


+def draw_rounded_rectangle(
+    scene: np.ndarray,
+    rect: Rect,
+    color: Color,
+    border_radius: int,
+) -> np.ndarray:
+    """
+    Draws a rounded rectangle on an image.
+
+    Parameters:
+        scene (np.ndarray): The image on which the rounded rectangle will be drawn.
+        rect (Rect): The rectangle to be drawn.
+        color (Color): The color of the rounded rectangle.
+        border_radius (int): The radius of the corner rounding.
+
+    Returns:
+        np.ndarray: The image with the rounded rectangle drawn on it.
+    """
+    x1, y1, x2, y2 = rect.as_xyxy_int_tuple()
+    width, height = x2 - x1, y2 - y1
+    border_radius = min(border_radius, min(width, height) // 2)
+
+    rectangle_coordinates = [
+        ((x1 + border_radius, y1), (x2 - border_radius, y2)),
+        ((x1, y1 + border_radius), (x2, y2 - border_radius)),
+    ]
+    circle_centers = [
+        (x1 + border_radius, y1 + border_radius),
+        (x2 - border_radius, y1 + border_radius),
+        (x1 + border_radius, y2 - border_radius),
+        (x2 - border_radius, y2 - border_radius),
+    ]
+
+    for coordinates in rectangle_coordinates:
+        cv2.rectangle(
+            img=scene,
+            pt1=coordinates[0],
+            pt2=coordinates[1],
+            color=color.as_bgr(),
+            thickness=-1,
+        )
+    for center in circle_centers:
+        cv2.circle(
+            img=scene,
+            center=center,
+            radius=border_radius,
+            color=color.as_bgr(),
+            thickness=-1,
+        )
+    return scene
+
+
 def draw_polygon(
    scene: np.ndarray, polygon: np.ndarray, color: Color, thickness: int = 2
 ) -> np.ndarray:
@@ -238,13 +290,13 @@ def draw_image(
    return scene


-def calculate_dynamic_text_scale(resolution_wh: Tuple[int, int]) -> float:
+def calculate_optimal_text_scale(resolution_wh: Tuple[int, int]) -> float:
    """
-    Calculate a dynamic font scale based on the resolution of an image.
+    Calculate font scale based on the resolution of an image.

    Parameters:
         resolution_wh (Tuple[int, int]): A tuple representing the width and height
-                 of the image.
+             of the image.

    Returns:
         float: The calculated font scale factor.
@@ -252,25 +304,17 @@ def calculate_dynamic_text_scale(resolution_wh: Tuple[int, int]) -> float:
    return min(resolution_wh) * 1e-3


-def calculate_dynamic_line_thickness(resolution_wh: Tuple[int, int]) -> int:
+def calculate_optimal_line_thickness(resolution_wh: Tuple[int, int]) -> int:
    """
-    Calculate a dynamic line thickness based on the resolution of an image.
+    Calculate line thickness based on the resolution of an image.

    Parameters:
        resolution_wh (Tuple[int, int]): A tuple representing the width and height
-                of the image.
+            of the image.

    Returns:
        int: The calculated line thickness in pixels.
    """
-    min_dimension = min(resolution_wh)
-    if min_dimension < 480:
+    if min(resolution_wh) < 1080:
        return 2
-    if min_dimension < 720:
-        return 2
-    if min_dimension < 1080:
-        return 2
-    if min_dimension < 2160:
-        return 4
-    else:
-        return 4
+    return 4
@@ -98,6 +98,11 @@ class Rect:
    width: float
    height: float

+    @classmethod
+    def from_xyxy(cls, xyxy: Tuple[float, float, float, float]) -> Rect:
+        x1, y1, x2, y2 = xyxy
+        return cls(x=x1, y=y1, width=x2 - x1, height=y2 - y1)
+
    @property
    def top_left(self) -> Point:
        return Point(x=self.x, y=self.y)
@@ -113,3 +118,11 @@ class Rect:
            width=self.width + 2 * padding,
            height=self.height + 2 * padding,
        )
+
+    def as_xyxy_int_tuple(self) -> Tuple[int, int, int, int]:
+        return (
+            int(self.x),
+            int(self.y),
+            int(self.x + self.width),
+            int(self.y + self.height),
+        )
@@ -5,12 +5,8 @@ from supervision.geometry.core import Point

 def get_polygon_center(polygon: np.ndarray) -> Point:
    """
-    Calculate the center of a polygon.
-
-    This function takes in a polygon as a 2-dimensional numpy ndarray and
-    returns the center of the polygon as a Point object.
-    The center is calculated as the mean of the polygon's vertices along each axis,
-    and is rounded down to the nearest integer.
+    Calculate the center of a polygon. The center is calculated as the center
+    of the solid figure formed by the points of the polygon

    Parameters:
        polygon (np.ndarray): A 2-dimensional numpy ndarray representing the
@@ -22,13 +18,24 @@ def get_polygon_center(polygon: np.ndarray) -> Point:

    Examples:
        ```python
-        from supervision.geometry.utils import get_polygon_center
        import numpy as np
+        import supervision as sv

-        vertices = np.array([[0, 0], [0, 1], [1, 1], [1, 0]])
-        get_center(vertices)
-        Point(x=0.5, y=0.5)
+        polygon = np.array([[0, 0], [0, 2], [2, 2], [2, 0]])
+        sv.get_polygon_center(polygon=polygon)
+        # Point(x=1, y=1)
        ```
    """
-    center = np.mean(polygon, axis=0).astype(int)
+
+    # This is one of the 3 candidate algorithms considered for centroid calculation.
+    # For a more detailed discussion, see PR #1084 and commit eb33176
+
+    shift_polygon = np.roll(polygon, -1, axis=0)
+    signed_areas = np.cross(polygon, shift_polygon) / 2
+    if signed_areas.sum() == 0:
+        center = np.mean(polygon, axis=0).round()
+        return Point(x=center[0], y=center[1])
+    centroids = (polygon + shift_polygon) / 3.0
+    center = np.average(centroids, axis=0, weights=signed_areas).round()
+
    return Point(x=center[0], y=center[1])
@@ -0,0 +1,418 @@
+from abc import ABC, abstractmethod
+from logging import warn
+from typing import List, Optional, Tuple, Union
+
+import cv2
+import numpy as np
+
+from supervision import Rect, pad_boxes
+from supervision.annotators.base import ImageType
+from supervision.draw.color import Color
+from supervision.draw.utils import draw_rounded_rectangle
+from supervision.keypoint.core import KeyPoints
+from supervision.keypoint.skeletons import SKELETONS_BY_VERTEX_COUNT
+from supervision.utils.conversion import ensure_cv2_image_for_annotation
+
+
+class BaseKeyPointAnnotator(ABC):
+    @abstractmethod
+    def annotate(self, scene: ImageType, key_points: KeyPoints) -> ImageType:
+        pass
+
+
+class VertexAnnotator(BaseKeyPointAnnotator):
+    """
+    A class that specializes in drawing skeleton vertices on images. It uses
+    specified key points to determine the locations where the vertices should be
+    drawn.
+    """
+
+    def __init__(
+        self,
+        color: Color = Color.ROBOFLOW,
+        radius: int = 4,
+    ) -> None:
+        """
+        Args:
+            color (Color, optional): The color to use for annotating key points.
+            radius (int, optional): The radius of the circles used to represent the key
+                points.
+        """
+        self.color = color
+        self.radius = radius
+
+    @ensure_cv2_image_for_annotation
+    def annotate(self, scene: ImageType, key_points: KeyPoints) -> ImageType:
+        """
+        Annotates the given scene with skeleton vertices based on the provided key
+        points. It draws circles at each key point location.
+
+        Args:
+            scene (ImageType): The image where skeleton vertices will be drawn.
+                `ImageType` is a flexible type, accepting either `numpy.ndarray` or
+                `PIL.Image.Image`.
+            key_points (KeyPoints): A collection of key points where each key point
+                consists of x and y coordinates.
+
+        Returns:
+            The annotated image, matching the type of `scene` (`numpy.ndarray`
+                or `PIL.Image.Image`)
+
+        Example:
+            ```python
+            import supervision as sv
+
+            image = ...
+            key_points = sv.KeyPoints(...)
+
+            vertex_annotator = sv.VertexAnnotator(
+                color=sv.Color.GREEN,
+                radius=10
+            )
+            annotated_frame = vertex_annotator.annotate(
+                scene=image.copy(),
+                key_points=key_points
+            )
+            ```
+
+        ![vertex-annotator-example](https://media.roboflow.com/
+        supervision-annotator-examples/vertex-annotator-example.png)
+        """
+        if len(key_points) == 0:
+            return scene
+
+        for xy in key_points.xy:
+            for x, y in xy:
+                cv2.circle(
+                    img=scene,
+                    center=(int(x), int(y)),
+                    radius=self.radius,
+                    color=self.color.as_bgr(),
+                    thickness=-1,
+                )
+
+        return scene
+
+
+class EdgeAnnotator(BaseKeyPointAnnotator):
+    """
+    A class that specializes in drawing skeleton edges on images using specified key
+    points. It connects key points with lines to form the skeleton structure.
+    """
+
+    def __init__(
+        self,
+        color: Color = Color.ROBOFLOW,
+        thickness: int = 2,
+        edges: Optional[List[Tuple[int, int]]] = None,
+    ) -> None:
+        """
+        Args:
+            color (Color, optional): The color to use for the edges.
+            thickness (int, optional): The thickness of the edges.
+            edges (Optional[List[Tuple[int, int]]]): The edges to draw.
+                If set to `None`, will attempt to select automatically.
+        """
+        self.color = color
+        self.thickness = thickness
+        self.edges = edges
+
+    @ensure_cv2_image_for_annotation
+    def annotate(self, scene: ImageType, key_points: KeyPoints) -> ImageType:
+        """
+        Annotates the given scene by drawing lines between specified key points to form
+        edges.
+
+        Args:
+            scene (ImageType): The image where skeleton edges will be drawn. `ImageType`
+                is a flexible type, accepting either `numpy.ndarray` or
+                `PIL.Image.Image`.
+            key_points (KeyPoints): A collection of key points where each key point
+                consists of x and y coordinates.
+
+        Returns:
+            Returns:
+                The annotated image, matching the type of `scene` (`numpy.ndarray`
+                    or `PIL.Image.Image`)
+
+        Example:
+            ```python
+            import supervision as sv
+
+            image = ...
+            key_points = sv.KeyPoints(...)
+
+            edge_annotator = sv.EdgeAnnotator(
+                color=sv.Color.GREEN,
+                thickness=5
+            )
+            annotated_frame = edge_annotator.annotate(
+                scene=image.copy(),
+                key_points=key_points
+            )
+            ```
+
+        ![edge-annotator-example](https://media.roboflow.com/
+        supervision-annotator-examples/edge-annotator-example.png)
+        """
+        if len(key_points) == 0:
+            return scene
+
+        for xy in key_points.xy:
+            edges = self.edges
+            if not edges:
+                edges = SKELETONS_BY_VERTEX_COUNT.get(len(xy))
+            if not edges:
+                warn(f"No skeleton found with {len(xy)} vertices")
+                return scene
+
+            for class_a, class_b in edges:
+                xy_a = xy[class_a - 1]
+                xy_b = xy[class_b - 1]
+                missing_a = np.allclose(xy_a, 0)
+                missing_b = np.allclose(xy_b, 0)
+                if missing_a or missing_b:
+                    continue
+
+                cv2.line(
+                    img=scene,
+                    pt1=(int(xy_a[0]), int(xy_a[1])),
+                    pt2=(int(xy_b[0]), int(xy_b[1])),
+                    color=self.color.as_bgr(),
+                    thickness=self.thickness,
+                )
+
+        return scene
+
+
+class VertexLabelAnnotator:
+    """
+    A class that draws labels of skeleton vertices on images. It uses specified key
+    points to determine the locations where the vertices should be drawn.
+    """
+
+    def __init__(
+        self,
+        color: Union[Color, List[Color]] = Color.ROBOFLOW,
+        text_color: Color = Color.WHITE,
+        text_scale: float = 0.5,
+        text_thickness: int = 1,
+        text_padding: int = 10,
+        border_radius: int = 0,
+    ):
+        """
+        Args:
+            color (Union[Color, List[Color]], optional): The color to use for each
+                keypoint label. If a list is provided, the colors will be used in order
+                for each keypoint.
+            text_color (Color, optional): The color to use for the labels.
+            text_scale (float, optional): The scale of the text.
+            text_thickness (int, optional): The thickness of the text.
+            text_padding (int, optional): The padding around the text.
+            border_radius (int, optional): The radius of the rounded corners of the
+                boxes. Set to a high value to produce circles.
+        """
+        self.border_radius: int = border_radius
+        self.color: Union[Color, List[Color]] = color
+        self.text_color: Color = text_color
+        self.text_scale: float = text_scale
+        self.text_thickness: int = text_thickness
+        self.text_padding: int = text_padding
+
+    def annotate(
+        self, scene: ImageType, key_points: KeyPoints, labels: List[str] = None
+    ) -> ImageType:
+        """
+        A class that draws labels of skeleton vertices on images. It uses specified key
+            points to determine the locations where the vertices should be drawn.
+
+        Args:
+            scene (ImageType): The image where vertex labels will be drawn. `ImageType`
+                is a flexible type, accepting either `numpy.ndarray` or
+                `PIL.Image.Image`.
+            key_points (KeyPoints): A collection of key points where each key point
+                consists of x and y coordinates.
+            labels (List[str], optional): A list of labels to be displayed on the
+                annotated image. If not provided, keypoint indices will be used.
+
+        Returns:
+            The annotated image, matching the type of `scene` (`numpy.ndarray`
+                or `PIL.Image.Image`)
+
+        Example:
+            ```python
+            import supervision as sv
+
+            image = ...
+            key_points = sv.KeyPoints(...)
+
+            vertex_label_annotator = sv.VertexLabelAnnotator(
+                color=sv.Color.GREEN,
+                text_color=sv.Color.BLACK,
+                border_radius=5
+            )
+            annotated_frame = vertex_label_annotator.annotate(
+                scene=image.copy(),
+                key_points=key_points
+            )
+            ```
+
+        ![vertex-label-annotator-example](https://media.roboflow.com/
+        supervision-annotator-examples/vertex-label-annotator-example.png)
+
+        !!! tip
+
+            `VertexLabelAnnotator` allows to customize the color of each keypoint label
+            values.
+
+        Example:
+            ```python
+            import supervision as sv
+
+            image = ...
+            key_points = sv.KeyPoints(...)
+
+            LABELS = [
+                "nose", "left eye", "right eye", "left ear",
+                "right ear", "left shoulder", "right shoulder", "left elbow",
+                "right elbow", "left wrist", "right wrist", "left hip",
+                "right hip", "left knee", "right knee", "left ankle",
+                "right ankle"
+            ]
+
+            COLORS = [
+                "#FF6347", "#FF6347", "#FF6347", "#FF6347",
+                "#FF6347", "#FF1493", "#00FF00", "#FF1493",
+                "#00FF00", "#FF1493", "#00FF00", "#FFD700",
+                "#00BFFF", "#FFD700", "#00BFFF", "#FFD700",
+                "#00BFFF"
+            ]
+            COLORS = [sv.Color.from_hex(color_hex=c) for c in COLORS]
+
+            vertex_label_annotator = sv.VertexLabelAnnotator(
+                color=COLORS,
+                text_color=sv.Color.BLACK,
+                border_radius=5
+            )
+            annotated_frame = vertex_label_annotator.annotate(
+                scene=image.copy(),
+                key_points=key_points,
+                labels=labels
+            )
+            ```
+        ![vertex-label-annotator-custom-example](https://media.roboflow.com/
+        supervision-annotator-examples/vertex-label-annotator-custom-example.png)
+        """
+        font = cv2.FONT_HERSHEY_SIMPLEX
+
+        skeletons_count, points_count, _ = key_points.xy.shape
+        if skeletons_count == 0:
+            return scene
+
+        anchors = key_points.xy.reshape(points_count * skeletons_count, 2).astype(int)
+        mask = np.all(anchors != 0, axis=1)
+
+        if not np.any(mask):
+            return scene
+
+        colors = self.preprocess_and_validate_colors(
+            colors=self.color,
+            points_count=points_count,
+            skeletons_count=skeletons_count,
+        )
+
+        labels = self.preprocess_and_validate_labels(
+            labels=labels, points_count=points_count, skeletons_count=skeletons_count
+        )
+
+        anchors = anchors[mask]
+        colors = colors[mask]
+        labels = labels[mask]
+
+        xyxy = np.array(
+            [
+                self.get_text_bounding_box(
+                    text=label,
+                    font=font,
+                    text_scale=self.text_scale,
+                    text_thickness=self.text_thickness,
+                    center_coordinates=tuple(anchor),
+                )
+                for anchor, label in zip(anchors, labels)
+            ]
+        )
+
+        xyxy_padded = pad_boxes(xyxy=xyxy, px=self.text_padding)
+
+        for text, color, box, box_padded in zip(labels, colors, xyxy, xyxy_padded):
+            draw_rounded_rectangle(
+                scene=scene,
+                rect=Rect.from_xyxy(box_padded),
+                color=color,
+                border_radius=self.border_radius,
+            )
+            cv2.putText(
+                img=scene,
+                text=text,
+                org=(box[0], box[1] + self.text_padding),
+                fontFace=font,
+                fontScale=self.text_scale,
+                color=self.text_color.as_rgb(),
+                thickness=self.text_thickness,
+                lineType=cv2.LINE_AA,
+            )
+
+        return scene
+
+    @staticmethod
+    def get_text_bounding_box(
+        text: str,
+        font: int,
+        text_scale: float,
+        text_thickness: int,
+        center_coordinates: Tuple[int, int],
+    ) -> Tuple[int, int, int, int]:
+        text_w, text_h = cv2.getTextSize(
+            text=text,
+            fontFace=font,
+            fontScale=text_scale,
+            thickness=text_thickness,
+        )[0]
+        center_x, center_y = center_coordinates
+        return (
+            center_x - text_w // 2,
+            center_y - text_h // 2,
+            center_x + text_w // 2,
+            center_y + text_h // 2,
+        )
+
+    @staticmethod
+    def preprocess_and_validate_labels(
+        labels: Optional[List[str]], points_count: int, skeletons_count: int
+    ) -> np.array:
+        if labels and len(labels) != points_count:
+            raise ValueError(
+                f"Number of labels ({len(labels)}) must match number of key points "
+                f"({points_count})."
+            )
+        if labels is None:
+            labels = [str(i) for i in range(points_count)]
+
+        return np.array(labels * skeletons_count)
+
+    @staticmethod
+    def preprocess_and_validate_colors(
+        colors: Optional[Union[Color, List[Color]]],
+        points_count: int,
+        skeletons_count: int,
+    ) -> np.array:
+        if isinstance(colors, list) and len(colors) != points_count:
+            raise ValueError(
+                f"Number of colors ({len(colors)}) must match number of key points "
+                f"({points_count})."
+            )
+        return (
+            np.array(colors * skeletons_count)
+            if isinstance(colors, list)
+            else np.array([colors] * points_count * skeletons_count)
+        )
@@ -0,0 +1,519 @@
+from __future__ import annotations
+
+from contextlib import suppress
+from dataclasses import dataclass, field
+from typing import Any, Dict, Iterator, List, Optional, Tuple, Union
+
+import numpy as np
+import numpy.typing as npt
+
+from supervision.config import CLASS_NAME_DATA_FIELD
+from supervision.detection.utils import get_data_item, is_data_equal
+from supervision.validators import validate_keypoints_fields
+
+
+@dataclass
+class KeyPoints:
+    """
+    The `sv.KeyPoints` class in the Supervision library standardizes results from
+    various keypoint detection and pose estimation models into a consistent format. This
+    class simplifies data manipulation and filtering, providing a uniform API for
+    integration with Supervision [keypoints annotators](/keypoint/annotators).
+
+    === "Ultralytics"
+
+        Use [`sv.KeyPoints.from_ultralytics`](/keypoint/core/#supervision.keypoint.core.KeyPoints.from_ultralytics)
+        method, which accepts [YOLOv8](https://github.com/ultralytics/ultralytics)
+        pose result.
+
+        ```python
+        import cv2
+        import supervision as sv
+        from ultralytics import YOLO
+
+        image = cv2.imread(<SOURCE_IMAGE_PATH>)
+        model = YOLO('yolov8s-pose.pt')
+
+        result = model(image)[0]
+        key_points = sv.KeyPoints.from_ultralytics(result)
+        ```
+
+    === "Inference"
+
+        Use [`sv.KeyPoints.from_inference`](/keypoint/core/#supervision.keypoint.core.KeyPoints.from_inference)
+        method, which accepts [Inference](https://inference.roboflow.com/) pose result.
+
+        ```python
+        import cv2
+        import supervision as sv
+        from inference import get_model
+
+        image = cv2.imread(<SOURCE_IMAGE_PATH>)
+        model = get_model(model_id=<POSE_MODEL_ID>, api_key=<ROBOFLOW_API_KEY>)
+
+        result = model.infer(image)[0]
+        key_points = sv.KeyPoints.from_inference(result)
+        ```
+
+    === "MediaPipe"
+
+        Use [`sv.KeyPoints.from_mediapipe`](/keypoint/core/#supervision.keypoint.core.KeyPoints.from_mediapipe)
+        method, which accepts [MediaPipe](https://github.com/google-ai-edge/mediapipe)
+        pose result.
+
+        ```python
+        import cv2
+        import mediapipe as mp
+        import supervision as sv
+
+        image = cv2.imread(<SOURCE_IMAGE_PATH>)
+        image_height, image_width, _ = image.shape
+        mediapipe_image = mp.Image(
+            image_format=mp.ImageFormat.SRGB,
+            data=cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
+
+        options = mp.tasks.vision.PoseLandmarkerOptions(
+            base_options=mp.tasks.BaseOptions(
+                model_asset_path="pose_landmarker_heavy.task"
+            ),
+            running_mode=mp.tasks.vision.RunningMode.IMAGE,
+            num_poses=2)
+
+        PoseLandmarker = mp.tasks.vision.PoseLandmarker
+        with PoseLandmarker.create_from_options(options) as landmarker:
+            pose_landmarker_result = landmarker.detect(mediapipe_image)
+
+        key_points = sv.KeyPoints.from_mediapipe(
+            pose_landmarker_result, (image_width, image_height))
+        ```
+
+    Attributes:
+        xy (np.ndarray): An array of shape `(n, 2)` containing
+            the bounding boxes coordinates in format `[x1, y1]`
+        confidence (Optional[np.ndarray]): An array of shape
+            `(n,)` containing the confidence scores of the keypoint keypoints.
+        class_id (Optional[np.ndarray]): An array of shape
+            `(n,)` containing the class ids of the keypoint keypoints.
+        data (Dict[str, Union[np.ndarray, List]]): A dictionary containing additional
+            data where each key is a string representing the data type, and the value
+            is either a NumPy array or a list of corresponding data.
+    """  # noqa: E501 // docs
+
+    xy: npt.NDArray[np.float32]
+    class_id: Optional[npt.NDArray[np.int_]] = None
+    confidence: Optional[npt.NDArray[np.float32]] = None
+    data: Dict[str, Union[npt.NDArray[Any], List]] = field(default_factory=dict)
+
+    def __post_init__(self):
+        validate_keypoints_fields(
+            xy=self.xy,
+            confidence=self.confidence,
+            class_id=self.class_id,
+            data=self.data,
+        )
+
+    def __len__(self) -> int:
+        """
+        Returns the number of keypoints in the `sv.KeyPoints` object.
+        """
+        return len(self.xy)
+
+    def __iter__(
+        self,
+    ) -> Iterator[
+        Tuple[
+            np.ndarray,
+            Optional[np.ndarray],
+            Optional[float],
+            Optional[int],
+            Optional[int],
+            Dict[str, Union[np.ndarray, List]],
+        ]
+    ]:
+        """
+        Iterates over the Keypoint object and yield a tuple of
+        `(xy, confidence, class_id, data)` for each keypoint detection.
+        """
+        for i in range(len(self.xy)):
+            yield (
+                self.xy[i],
+                self.confidence[i] if self.confidence is not None else None,
+                self.class_id[i] if self.class_id is not None else None,
+                get_data_item(self.data, i),
+            )
+
+    def __eq__(self, other: KeyPoints) -> bool:
+        return all(
+            [
+                np.array_equal(self.xy, other.xy),
+                np.array_equal(self.class_id, other.class_id),
+                np.array_equal(self.confidence, other.confidence),
+                is_data_equal(self.data, other.data),
+            ]
+        )
+
+    @classmethod
+    def from_inference(cls, inference_result: Union[dict, Any]) -> KeyPoints:
+        """
+        Create a `sv.KeyPoints` object from the [Roboflow](https://roboflow.com/)
+        API inference result or the [Inference](https://inference.roboflow.com/)
+        package results.
+
+        Args:
+            inference_result (dict, any): The result from the
+                Roboflow API or Inference package containing predictions with keypoints.
+
+        Returns:
+            A `sv.KeyPoints` object containing the keypoint coordinates, class IDs,
+                and class names, and confidences of each keypoint.
+
+        Examples:
+            ```python
+            import cv2
+            import supervision as sv
+            from inference import get_model
+
+            image = cv2.imread(<SOURCE_IMAGE_PATH>)
+            model = get_model(model_id=<POSE_MODEL_ID>, api_key=<ROBOFLOW_API_KEY>)
+
+            result = model.infer(image)[0]
+            key_points = sv.KeyPoints.from_inference(result)
+            ```
+
+            ```python
+            import cv2
+            import supervision as sv
+            from inference_sdk import InferenceHTTPClient
+
+            image = cv2.imread(<SOURCE_IMAGE_PATH>)
+            client = InferenceHTTPClient(
+                api_url="https://detect.roboflow.com",
+                api_key=<ROBOFLOW_API_KEY>
+            )
+
+            result = client.infer(image, model_id=<POSE_MODEL_ID>)
+            key_points = sv.KeyPoints.from_inference(result)
+            ```
+        """
+        if isinstance(inference_result, list):
+            raise ValueError(
+                "from_inference() operates on a single result at a time."
+                "You can retrieve it like so:  inference_result = model.infer(image)[0]"
+            )
+
+        with suppress(AttributeError):
+            inference_result = inference_result.dict(exclude_none=True, by_alias=True)
+
+        if not inference_result.get("predictions"):
+            return cls.empty()
+
+        xy = []
+        confidence = []
+        class_id = []
+        class_names = []
+
+        for prediction in inference_result["predictions"]:
+            prediction_xy = []
+            prediction_confidence = []
+            for keypoint in prediction["keypoints"]:
+                prediction_xy.append([keypoint["x"], keypoint["y"]])
+                prediction_confidence.append(keypoint["confidence"])
+            xy.append(prediction_xy)
+            confidence.append(prediction_confidence)
+
+            class_id.append(prediction["class_id"])
+            class_names.append(prediction["class"])
+
+        data = {CLASS_NAME_DATA_FIELD: np.array(class_names)}
+
+        return cls(
+            xy=np.array(xy, dtype=np.float32),
+            confidence=np.array(confidence, dtype=np.float32),
+            class_id=np.array(class_id, dtype=int),
+            data=data,
+        )
+
+    @classmethod
+    def from_mediapipe(
+        cls, mediapipe_results, resolution_wh: Tuple[int, int]
+    ) -> KeyPoints:
+        """
+        Creates a `sv.KeyPoints` instance from a
+        [MediaPipe](https://github.com/google-ai-edge/mediapipe)
+        pose landmark detection inference result.
+
+        Args:
+            mediapipe_results (Union[PoseLandmarkerResult, SolutionOutputs]):
+                The output results from Mediapipe. It supports both: the inference
+                result `PoseLandmarker` and the legacy one from `Pose`.
+            resolution_wh (Tuple[int, int]): A tuple of the form `(width, height)`
+                representing the resolution of the frame.
+
+        Returns:
+            A `sv.KeyPoints` object containing the keypoint coordinates and
+                confidences of each keypoint.
+
+        !!! tip
+            Before you start, download model bundles from the
+            [MediaPipe website](https://ai.google.dev/edge/mediapipe/solutions/vision/pose_landmarker/index#models).
+
+        Examples:
+            ```python
+            import cv2
+            import mediapipe as mp
+            import supervision as sv
+
+            image = cv2.imread(<SOURCE_IMAGE_PATH>)
+            image_height, image_width, _ = image.shape
+            mediapipe_image = mp.Image(
+                image_format=mp.ImageFormat.SRGB,
+                data=cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
+
+            options = mp.tasks.vision.PoseLandmarkerOptions(
+                base_options=mp.tasks.BaseOptions(
+                    model_asset_path="pose_landmarker_heavy.task"
+                ),
+                running_mode=mp.tasks.vision.RunningMode.IMAGE,
+                num_poses=2)
+
+            PoseLandmarker = mp.tasks.vision.PoseLandmarker
+            with PoseLandmarker.create_from_options(options) as landmarker:
+                pose_landmarker_result = landmarker.detect(mediapipe_image)
+
+            key_points = sv.KeyPoints.from_mediapipe(
+                pose_landmarker_result, (image_width, image_height))
+            ```
+        """  # noqa: E501 // docs
+        results = mediapipe_results.pose_landmarks
+        if not isinstance(mediapipe_results.pose_landmarks, list):
+            if mediapipe_results.pose_landmarks is None:
+                results = []
+            else:
+                results = [
+                    [landmark for landmark in mediapipe_results.pose_landmarks.landmark]
+                ]
+
+        if len(results) == 0:
+            return cls.empty()
+
+        xy = []
+        confidence = []
+        for pose in results:
+            prediction_xy = []
+            prediction_confidence = []
+            for landmark in pose:
+                keypoint_xy = [
+                    landmark.x * resolution_wh[0],
+                    landmark.y * resolution_wh[1],
+                ]
+                prediction_xy.append(keypoint_xy)
+                prediction_confidence.append(landmark.visibility)
+
+            xy.append(prediction_xy)
+            confidence.append(prediction_confidence)
+
+        return cls(
+            xy=np.array(xy, dtype=np.float32),
+            confidence=np.array(confidence, dtype=np.float32),
+        )
+
+    @classmethod
+    def from_ultralytics(cls, ultralytics_results) -> KeyPoints:
+        """
+        Creates a `sv.KeyPoints` instance from a
+        [YOLOv8](https://github.com/ultralytics/ultralytics) pose inference result.
+
+        Args:
+            ultralytics_results (ultralytics.engine.results.Keypoints):
+                The output Results instance from YOLOv8
+
+        Returns:
+            A `sv.KeyPoints` object containing the keypoint coordinates, class IDs,
+                and class names, and confidences of each keypoint.
+
+        Examples:
+            ```python
+            import cv2
+            import supervision as sv
+            from ultralytics import YOLO
+
+            image = cv2.imread(<SOURCE_IMAGE_PATH>)
+            model = YOLO('yolov8s-pose.pt')
+
+            result = model(image)[0]
+            key_points = sv.KeyPoints.from_ultralytics(result)
+            ```
+        """
+        if ultralytics_results.keypoints.xy.numel() == 0:
+            return cls.empty()
+
+        xy = ultralytics_results.keypoints.xy.cpu().numpy()
+        class_id = ultralytics_results.boxes.cls.cpu().numpy().astype(int)
+        class_names = np.array([ultralytics_results.names[i] for i in class_id])
+
+        confidence = ultralytics_results.keypoints.conf.cpu().numpy()
+        data = {CLASS_NAME_DATA_FIELD: class_names}
+        return cls(xy, class_id, confidence, data)
+
+    @classmethod
+    def from_yolo_nas(cls, yolo_nas_results) -> KeyPoints:
+        """
+        Create a `sv.KeyPoints` instance from a [YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS-POSE.md)
+        pose inference results.
+
+        Args:
+            yolo_nas_results (ImagePoseEstimationPrediction): The output object from
+                YOLO NAS.
+
+        Returns:
+            A `sv.KeyPoints` object containing the keypoint coordinates, class IDs,
+                and class names, and confidences of each keypoint.
+
+        Examples:
+            ```python
+            import cv2
+            import torch
+            import supervision as sv
+            import super_gradients
+
+            image = cv2.imread(<SOURCE_IMAGE_PATH>)
+
+            device = "cuda" if torch.cuda.is_available() else "cpu"
+            model = super_gradients.training.models.get(
+                "yolo_nas_pose_s", pretrained_weights="coco_pose").to(device)
+
+            results = model.predict(image, conf=0.1)
+            key_points = sv.KeyPoints.from_yolo_nas(results)
+            ```
+        """  # noqa: E501 // docs
+        if len(yolo_nas_results.prediction.poses) == 0:
+            return cls.empty()
+
+        xy = yolo_nas_results.prediction.poses[:, :, :2]
+        confidence = yolo_nas_results.prediction.poses[:, :, 2]
+
+        # yolo_nas_results treats params differently.
+        # prediction.labels may not exist, whereas class_names might be None
+        if hasattr(yolo_nas_results.prediction, "labels"):
+            class_id = yolo_nas_results.prediction.labels  # np.array[int]
+        else:
+            class_id = None
+
+        data = {}
+        if class_id is not None and yolo_nas_results.class_names is not None:
+            class_names = []
+            for c_id in class_id:
+                name = yolo_nas_results.class_names[c_id]  # tuple[str]
+                class_names.append(name)
+            data[CLASS_NAME_DATA_FIELD] = class_names
+
+        return cls(
+            xy=xy,
+            confidence=confidence,
+            class_id=class_id,
+            data=data,
+        )
+
+    def __getitem__(
+        self, index: Union[int, slice, List[int], np.ndarray, str]
+    ) -> Union[KeyPoints, List, np.ndarray, None]:
+        """
+        Get a subset of the `sv.KeyPoints` object or access an item from its data field.
+
+        When provided with an integer, slice, list of integers, or a numpy array, this
+        method returns a new `sv.KeyPoints` object that represents a subset of the
+        original `sv.KeyPoints`. When provided with a string, it accesses the
+        corresponding item in the data dictionary.
+
+        Args:
+            index (Union[int, slice, List[int], np.ndarray, str]): The index, indices,
+                or key to access a subset of the `sv.KeyPoints` or an item from the
+                data.
+
+        Returns:
+            A subset of the `sv.KeyPoints` object or an item from the data field.
+
+        Examples:
+            ```python
+            import supervision as sv
+
+            key_points = sv.KeyPoints()
+
+            # access the first keypoint using an integer index
+            key_points[0]
+
+            # access the first 10 keypoints using index slice
+            key_points[0:10]
+
+            # access selected keypoints using a list of indices
+            key_points[[0, 2, 4]]
+
+            # access keypoints with selected class_id
+            key_points[key_points.class_id == 0]
+
+            # access keypoints with confidence greater than 0.5
+            key_points[key_points.confidence > 0.5]
+            ```
+        """
+        if isinstance(index, str):
+            return self.data.get(index)
+        if isinstance(index, int):
+            index = [index]
+        return KeyPoints(
+            xy=self.xy[index],
+            confidence=self.confidence[index] if self.confidence is not None else None,
+            class_id=self.class_id[index] if self.class_id is not None else None,
+            data=get_data_item(self.data, index),
+        )
+
+    def __setitem__(self, key: str, value: Union[np.ndarray, List]):
+        """
+        Set a value in the data dictionary of the `sv.KeyPoints` object.
+
+        Args:
+            key (str): The key in the data dictionary to set.
+            value (Union[np.ndarray, List]): The value to set for the key.
+
+        Examples:
+            ```python
+            import cv2
+            import supervision as sv
+            from ultralytics import YOLO
+
+            image = cv2.imread(<SOURCE_IMAGE_PATH>)
+            model = YOLO('yolov8s.pt')
+
+            result = model(image)[0]
+            keypoints = sv.KeyPoints.from_ultralytics(result)
+
+            keypoints['class_name'] = [
+                 model.model.names[class_id]
+                 for class_id
+                 in keypoints.class_id
+             ]
+            ```
+        """
+        if not isinstance(value, (np.ndarray, list)):
+            raise TypeError("Value must be a np.ndarray or a list")
+
+        if isinstance(value, list):
+            value = np.array(value)
+
+        self.data[key] = value
+
+    @classmethod
+    def empty(cls) -> KeyPoints:
+        """
+        Create an empty Keypoints object with no keypoints.
+
+        Returns:
+            An empty `sv.KeyPoints` object.
+
+        Examples:
+            ```python
+            import supervision as sv
+
+            key_points = sv.KeyPoints.empty()
+            ```
+        """
+        return cls(xy=np.empty((0, 0, 2), dtype=np.float32))
@@ -0,0 +1,74 @@
+from enum import Enum
+from typing import Dict, List, Tuple
+
+Edges = List[Tuple[int, int]]
+
+
+class Skeleton(Enum):
+    COCO = [
+        (1, 2),
+        (1, 3),
+        (2, 3),
+        (2, 4),
+        (3, 5),
+        (6, 12),
+        (6, 7),
+        (6, 8),
+        (7, 13),
+        (7, 9),
+        (8, 10),
+        (9, 11),
+        (12, 13),
+        (14, 12),
+        (15, 13),
+        (16, 14),
+        (17, 15),
+    ]
+
+    GHUM = [
+        (1, 2),
+        (1, 5),
+        (2, 3),
+        (3, 4),
+        (4, 8),
+        (5, 6),
+        (6, 7),
+        (7, 9),
+        (10, 11),
+        (12, 13),
+        (12, 14),
+        (12, 24),
+        (13, 15),
+        (13, 25),
+        (14, 16),
+        (15, 17),
+        (16, 18),
+        (15, 19),
+        (16, 22),
+        (17, 19),
+        (17, 21),
+        (17, 23),
+        (18, 20),
+        (19, 21),
+        (24, 25),
+        (24, 26),
+        (25, 27),
+        (26, 28),
+        (27, 29),
+        (28, 30),
+        (28, 32),
+        (29, 31),
+        (29, 33),
+        (30, 32),
+        (31, 33),
+    ]
+
+
+SKELETONS_BY_EDGE_COUNT: Dict[int, Edges] = {}
+SKELETONS_BY_VERTEX_COUNT: Dict[int, Edges] = {}
+
+for skeleton in Skeleton:
+    SKELETONS_BY_EDGE_COUNT[len(skeleton.value)] = skeleton.value
+
+    unique_vertices = set(vertex for edge in skeleton.value for vertex in edge)
+    SKELETONS_BY_VERTEX_COUNT[len(unique_vertices)] = skeleton.value
@@ -3,6 +3,7 @@ from typing import List, Tuple
 import numpy as np

 from supervision.detection.core import Detections
+from supervision.detection.utils import box_iou_batch
 from supervision.tracker.byte_tracker import matching
 from supervision.tracker.byte_tracker.basetrack import BaseTrack, TrackState
 from supervision.tracker.byte_tracker.kalman_filter import KalmanFilter
@@ -11,8 +12,9 @@ from supervision.utils.internal import deprecated_parameter

 class STrack(BaseTrack):
    shared_kalman = KalmanFilter()
+    _external_count = 0

-    def __init__(self, tlwh, score, class_ids):
+    def __init__(self, tlwh, score, class_ids, minimum_consecutive_frames):
        # wait activate
        self._tlwh = np.asarray(tlwh, dtype=np.float32)
        self.kalman_filter = None
@@ -23,6 +25,10 @@ class STrack(BaseTrack):
        self.class_ids = class_ids
        self.tracklet_len = 0

+        self.external_track_id = -1
+
+        self.minimum_consecutive_frames = minimum_consecutive_frames
+
    def predict(self):
        mean_state = self.mean.copy()
        if self.state != TrackState.Tracked:
@@ -52,7 +58,7 @@ class STrack(BaseTrack):
    def activate(self, kalman_filter, frame_id):
        """Start a new tracklet"""
        self.kalman_filter = kalman_filter
-        self.track_id = self.next_id()
+        self.internal_track_id = self.next_id()
        self.mean, self.covariance = self.kalman_filter.initiate(
            self.tlwh_to_xyah(self._tlwh)
        )
@@ -61,6 +67,10 @@ class STrack(BaseTrack):
        self.state = TrackState.Tracked
        if frame_id == 1:
            self.is_activated = True
+
+        if self.minimum_consecutive_frames == 1:
+            self.external_track_id = self.next_external_id()
+
        self.frame_id = frame_id
        self.start_frame = frame_id

@@ -70,10 +80,10 @@ class STrack(BaseTrack):
        )
        self.tracklet_len = 0
        self.state = TrackState.Tracked
-        self.is_activated = True
+
        self.frame_id = frame_id
        if new_id:
-            self.track_id = self.next_id()
+            self.internal_track_id = self.next_id()
        self.score = new_track.score

    def update(self, new_track, frame_id):
@@ -92,7 +102,10 @@ class STrack(BaseTrack):
            self.mean, self.covariance, self.tlwh_to_xyah(new_tlwh)
        )
        self.state = TrackState.Tracked
-        self.is_activated = True
+        if self.tracklet_len == self.minimum_consecutive_frames:
+            self.is_activated = True
+            if self.external_track_id == -1:
+                self.external_track_id = self.next_external_id()

        self.score = new_track.score

@@ -130,6 +143,15 @@ class STrack(BaseTrack):
    def to_xyah(self):
        return self.tlwh_to_xyah(self.tlwh)

+    @staticmethod
+    def next_external_id():
+        STrack._external_count += 1
+        return STrack._external_count
+
+    @staticmethod
+    def reset_external_counter():
+        STrack._external_count = 0
+
    @staticmethod
    def tlbr_to_tlwh(tlbr):
        ret = np.asarray(tlbr).copy()
@@ -143,7 +165,9 @@ class STrack(BaseTrack):
        return ret

    def __repr__(self):
-        return "OT_{}_({}-{})".format(self.track_id, self.start_frame, self.end_frame)
+        return "OT_{}_({}-{})".format(
+            self.internal_track_id, self.start_frame, self.end_frame
+        )


 def detections2boxes(detections: Detections) -> np.ndarray:
@@ -185,6 +209,10 @@ class ByteTrack:
            Increasing minimum_matching_threshold improves accuracy but risks fragmentation.
            Decreasing it improves completeness but risks false positives and drift.
        frame_rate (int, optional): The frame rate of the video.
+        minimum_consecutive_frames (int, optional): Number of consecutive frames that an object must
+            be tracked before it is considered a 'valid' track.
+            Increasing minimum_consecutive_frames prevents the creation of accidental tracks from
+            false detection or double detection, but risks missing shorter tracks.
    """  # noqa: E501 // docs

    @deprecated_parameter(
@@ -217,6 +245,7 @@ class ByteTrack:
        lost_track_buffer: int = 30,
        minimum_matching_threshold: float = 0.8,
        frame_rate: int = 30,
+        minimum_consecutive_frames: int = 1,
    ):
        self.track_activation_threshold = track_activation_threshold
        self.minimum_matching_threshold = minimum_matching_threshold
@@ -224,6 +253,7 @@ class ByteTrack:
        self.frame_id = 0
        self.det_thresh = self.track_activation_threshold + 0.1
        self.max_time_lost = int(frame_rate / 30.0 * lost_track_buffer)
+        self.minimum_consecutive_frames = minimum_consecutive_frames
        self.kalman_filter = KalmanFilter()

        self.tracked_tracks: List[STrack] = []
@@ -270,27 +300,31 @@ class ByteTrack:
            ```
        """

-        tracks = self.update_with_tensors(
-            tensors=detections2boxes(detections=detections)
-        )
-        detections = Detections.empty()
+        tensors = detections2boxes(detections=detections)
+        tracks = self.update_with_tensors(tensors=tensors)
+
        if len(tracks) > 0:
-            detections.xyxy = np.array(
-                [track.tlbr for track in tracks], dtype=np.float32
-            )
-            detections.class_id = np.array(
-                [int(t.class_ids) for t in tracks], dtype=int
-            )
-            detections.tracker_id = np.array(
-                [int(t.track_id) for t in tracks], dtype=int
-            )
-            detections.confidence = np.array(
-                [t.score for t in tracks], dtype=np.float32
-            )
+            detection_bounding_boxes = np.asarray([det[:4] for det in tensors])
+            track_bounding_boxes = np.asarray([track.tlbr for track in tracks])
+
+            ious = box_iou_batch(detection_bounding_boxes, track_bounding_boxes)
+
+            iou_costs = 1 - ious
+
+            matches, _, _ = matching.linear_assignment(iou_costs, 0.5)
+            detections.tracker_id = np.full(len(detections), -1, dtype=int)
+            for i_detection, i_track in matches:
+                detections.tracker_id[i_detection] = int(
+                    tracks[i_track].external_track_id
+                )
+
+            return detections[detections.tracker_id != -1]
+
        else:
+            detections = Detections.empty()
            detections.tracker_id = np.array([], dtype=int)

-        return detections
+            return detections

    def reset(self):
        """
@@ -306,6 +340,7 @@ class ByteTrack:
        self.lost_tracks: List[STrack] = []
        self.removed_tracks: List[STrack] = []
        BaseTrack.reset_counter()
+        STrack.reset_external_counter()

    def update_with_tensors(self, tensors: np.ndarray) -> List[STrack]:
        """
@@ -343,7 +378,7 @@ class ByteTrack:
        if len(dets) > 0:
            """Detections"""
            detections = [
-                STrack(STrack.tlbr_to_tlwh(tlbr), s, c)
+                STrack(STrack.tlbr_to_tlwh(tlbr), s, c, self.minimum_consecutive_frames)
                for (tlbr, s, c) in zip(dets, scores_keep, class_ids_keep)
            ]
        else:
@@ -385,7 +420,7 @@ class ByteTrack:
        if len(dets_second) > 0:
            """Detections"""
            detections_second = [
-                STrack(STrack.tlbr_to_tlwh(tlbr), s, c)
+                STrack(STrack.tlbr_to_tlwh(tlbr), s, c, self.minimum_consecutive_frames)
                for (tlbr, s, c) in zip(dets_second, scores_second, class_ids_second)
            ]
        else:
@@ -452,7 +487,7 @@ class ByteTrack:
        self.lost_tracks = sub_tracks(self.lost_tracks, self.tracked_tracks)
        self.lost_tracks.extend(lost_stracks)
        self.lost_tracks = sub_tracks(self.lost_tracks, self.removed_tracks)
-        self.removed_tracks.extend(removed_stracks)
+        self.removed_tracks = removed_stracks
        self.tracked_tracks, self.lost_tracks = remove_duplicate_tracks(
            self.tracked_tracks, self.lost_tracks
        )
@@ -466,22 +501,22 @@ def joint_tracks(
 ) -> List[STrack]:
    """
    Joins two lists of tracks, ensuring that the resulting list does not
-    contain tracks with duplicate track_id values.
+    contain tracks with duplicate internal_track_id values.

    Parameters:
-        track_list_a: First list of tracks (with track_id attribute).
-        track_list_b: Second list of tracks (with track_id attribute).
+        track_list_a: First list of tracks (with internal_track_id attribute).
+        track_list_b: Second list of tracks (with internal_track_id attribute).

    Returns:
        Combined list of tracks from track_list_a and track_list_b
-            without duplicate track_id values.
+            without duplicate internal_track_id values.
    """
    seen_track_ids = set()
    result = []

    for track in track_list_a + track_list_b:
-        if track.track_id not in seen_track_ids:
-            seen_track_ids.add(track.track_id)
+        if track.internal_track_id not in seen_track_ids:
+            seen_track_ids.add(track.internal_track_id)
            result.append(track)

    return result
@@ -490,17 +525,17 @@ def joint_tracks(
 def sub_tracks(track_list_a: List, track_list_b: List) -> List[int]:
    """
    Returns a list of tracks from track_list_a after removing any tracks
-    that share the same track_id with tracks in track_list_b.
+    that share the same internal_track_id with tracks in track_list_b.

    Parameters:
-        track_list_a: List of tracks (with track_id attribute).
-        track_list_b: List of tracks (with track_id attribute) to
+        track_list_a: List of tracks (with internal_track_id attribute).
+        track_list_b: List of tracks (with internal_track_id attribute) to
            be subtracted from track_list_a.
    Returns:
        List of remaining tracks from track_list_a after subtraction.
    """
-    tracks = {track.track_id: track for track in track_list_a}
-    track_ids_b = {track.track_id for track in track_list_b}
+    tracks = {track.internal_track_id: track for track in track_list_a}
+    track_ids_b = {track.internal_track_id for track in track_list_b}

    for track_id in track_ids_b:
        tracks.pop(track_id, None)
@@ -0,0 +1,132 @@
+from functools import wraps
+from typing import List
+
+import cv2
+import numpy as np
+from PIL import Image
+
+from supervision.annotators.base import ImageType
+
+
+def ensure_cv2_image_for_annotation(annotate_func):
+    """
+    Decorates `BaseAnnotator.annotate` implementations, converts scene to
+    an image type used internally by the annotators, converts back when annotation
+    is complete.
+
+    Assumes the annotators modify the scene in-place.
+    """
+
+    @wraps(annotate_func)
+    def wrapper(self, scene: ImageType, *args, **kwargs):
+        if isinstance(scene, np.ndarray):
+            return annotate_func(self, scene, *args, **kwargs)
+
+        if isinstance(scene, Image.Image):
+            scene_np = pillow_to_cv2(scene)
+            annotated_np = annotate_func(self, scene_np, *args, **kwargs)
+            scene.paste(cv2_to_pillow(annotated_np))
+            return scene
+
+        raise ValueError(f"Unsupported image type: {type(scene)}")
+
+    return wrapper
+
+
+def ensure_cv2_image_for_processing(image_processing_fun):
+    """
+    Decorates image processing functions that accept np.ndarray, converting `image` to
+    np.ndarray, converts back when processing is complete.
+
+    Assumes the annotators do NOT modify the scene in-place.
+    """
+
+    @wraps(image_processing_fun)
+    def wrapper(image: ImageType, *args, **kwargs):
+        if isinstance(image, np.ndarray):
+            return image_processing_fun(image, *args, **kwargs)
+
+        if isinstance(image, Image.Image):
+            scene = pillow_to_cv2(image)
+            annotated = image_processing_fun(scene, *args, **kwargs)
+            return cv2_to_pillow(annotated)
+
+        raise ValueError(f"Unsupported image type: {type(image)}")
+
+    return wrapper
+
+
+def ensure_pil_image_for_annotation(annotate_func):
+    """
+    Decorates image processing functions that accept np.ndarray, converting `image` to
+    PIL image, converts back when processing is complete.
+
+    Assumes the annotators modify the scene in-place.
+    """
+
+    @wraps(annotate_func)
+    def wrapper(self, scene: ImageType, *args, **kwargs):
+        if isinstance(scene, np.ndarray):
+            scene_pil = cv2_to_pillow(scene)
+            annotated_pil = annotate_func(self, scene_pil, *args, **kwargs)
+            np.copyto(scene, pillow_to_cv2(annotated_pil))
+            return scene
+
+        if isinstance(scene, Image.Image):
+            return annotate_func(self, scene, *args, **kwargs)
+
+        raise ValueError(f"Unsupported image type: {type(scene)}")
+
+    return wrapper
+
+
+def images_to_cv2(images: List[ImageType]) -> List[np.ndarray]:
+    """
+    Converts images provided either as Pillow images or OpenCV
+    images into OpenCV format.
+
+    Args:
+        images (List[ImageType]): Images to be converted
+
+    Returns:
+        List[np.ndarray]: List of input images in OpenCV format
+            (with order preserved).
+
+    """
+    result = []
+    for image in images:
+        if issubclass(type(image), Image.Image):
+            image = pillow_to_cv2(image)
+        result.append(image)
+    return result
+
+
+def pillow_to_cv2(image: Image.Image) -> np.ndarray:
+    """
+    Converts Pillow image into OpenCV image, handling RGB -> BGR
+    conversion.
+
+    Args:
+        image (Image.Image): Pillow image (in RGB format).
+
+    Returns:
+        (np.ndarray): Input image converted to OpenCV format.
+    """
+    scene = np.array(image)
+    scene = cv2.cvtColor(scene, cv2.COLOR_RGB2BGR)
+    return scene
+
+
+def cv2_to_pillow(image: np.ndarray) -> Image.Image:
+    """
+    Converts OpenCV image into Pillow image, handling BGR -> RGB
+    conversion.
+
+    Args:
+        image (np.ndarray): OpenCV image (in BGR format).
+
+    Returns:
+        (Image.Image): Input image converted to Pillow format.
+    """
+    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+    return Image.fromarray(image)
--- a/Show More
+++ b/Show More