supervision-0.13.0 release

`supervision-0.13.0` release

.gitattributes

@@ -0,0 +1 @@
+*.ipynb linguist-vendored

.github/workflows/docs.yml

@@ -5,14 +5,15 @@ on:
     branches:
       - master
       - main
+      - develop
 jobs:
   deploy:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
-      - uses: actions/setup-python@v2
+      - uses: actions/checkout@v3
+      - uses: actions/setup-python@v4
         with:
           python-version: 3.x
       - run: pip install mkdocs-material
       - run: pip install "mkdocstrings[python]"
-      - run: mkdocs gh-deploy --force
+      - run: mkdocs gh-deploy --force

.github/workflows/publish.yml

@@ -16,17 +16,32 @@ jobs:
         with:
           ref: ${{ github.head_ref }}
       - name: 🐍 Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v2
+        uses: actions/setup-python@v4
         with:
           python-version: ${{ matrix.python-version }}
       - name: 🦾 Install dependencies
         run: |
+          python -m pip install --upgrade virtualenv
           python -m pip install --upgrade pip
-          pip install ".[dev]"
-      - name: 🚀 Publish to PyPi
-        env: 
-          PYPI_USERNAME: ${{ secrets.PYPI_USERNAME }}
-          PYPI_PASSWORD: ${{ secrets.PYPI_PASSWORD }}
-          PYPI_TEST_PASSWORD: ${{ secrets.PYPI_TEST_PASSWORD }}
+          virtualenv venv
+          source venv/bin/activate
+          python -m pip install --upgrade pip
+          python -m pip install --upgrade poetry
+          poetry install
+
+      - name:  🏗️ Build source and wheel distributions
         run: |
-          make publish -e PYPI_USERNAME=$PYPI_USERNAME -e PYPI_PASSWORD=$PYPI_PASSWORD -e PYPI_TEST_PASSWORD=$PYPI_TEST_PASSWORD
+          python -m pip install --upgrade build twine
+          python -m build
+          twine check --strict dist/*
+      - name: 🚀 Publish to PyPi
+        uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          user: ${{ secrets.PYPI_USERNAME }}
+          password: ${{ secrets.PYPI_PASSWORD }}
+      - name: 🚀 Publish to Test-PyPi
+        uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          repository-url: https://test.pypi.org/legacy/
+          user: ${{ secrets.PYPI_USERNAME }}
+          password: ${{ secrets.PYPI_TEST_PASSWORD }}

.github/workflows/test.yml

@@ -2,21 +2,19 @@ name: Test WorkFlow
 
 on:
   pull_request:
-    branches: [main]
+    branches: [main, develop]
 
 jobs:
   build:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: [3.7, 3.8, 3.9, '3.10']
+        python-version: ["3.8", "3.9", "3.10","3.11"]
     steps:
       - name: 🛎️ Checkout
         uses: actions/checkout@v3
-        with:
-          ref: ${{ github.head_ref }}
       - name: 🐍 Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v2
+        uses: actions/setup-python@v4
         with:
           python-version: ${{ matrix.python-version }}
       - name: 🦾 Install dependencies

CITATION.cff

@@ -2,7 +2,7 @@ cff-version: 1.2.0
 author: Roboflow
 message: If you use this software, please cite it as below.
 title: supervision
-version: 0.1.0
+version: 0.11.1
 date-released: 2023-01-19
 license: MIT
 repository-code: https://github.com/roboflow/supervision

CONTRIBUTING.md

@@ -24,7 +24,7 @@ Before you contribute a new feature, consider submitting an Issue to discuss the
 
 ## How to Contribute Changes
 
-First, fork this repository to your own GitHub account. Create a new branch that describes your changes (i.e. `line-counter-docs`). Push your changes to the branch on your fork and then submit a pull request to this repository.
+First, fork this repository to your own GitHub account. Create a new branch that describes your changes (i.e. `line-counter-docs`). Push your changes to the branch on your fork and then submit a pull request to `develop` branch of this repository.
 
 When creating new functions, please ensure you have the following:
 
@@ -32,6 +32,7 @@ When creating new functions, please ensure you have the following:
 2. Unit tests for the function.
 3. Examples in the documentation for the function.
 4. Created an entry in our docs to autogenerate the documentation for the function.
+5. Please share google colab with minimal code to test new feature or reproduce PR whenever it is possible. Please ensure that google colab can be accessed without any issue. 
 
 All pull requests will be reviewed by the maintainers of the project. We will provide feedback and ask for changes if necessary.
 
@@ -48,4 +49,4 @@ So far, **there is no types checking with mypy**. See [issue](https://github.com
 
 ## 🧪 tests 
 
-[`pytests`](https://docs.pytest.org/en/7.1.x/) is used to run our tests.
+[`pytests`](https://docs.pytest.org/en/7.1.x/) is used to run our tests.

Makefile

@@ -16,7 +16,7 @@ check_code_quality:
 	flake8 $(check_dirs) --count --max-line-length=88 --exit-zero  --ignore=D --extend-ignore=E203,E501,W503  --statistics
 	
 publish:
-	python setup.py sdist bdist_wheel
+	poetry build
 	twine upload -r testpypi dist/* -u ${PYPI_USERNAME} -p ${PYPI_TEST_PASSWORD} --verbose 
 	twine check dist/*
 	twine upload dist/* -u ${PYPI_USERNAME} -p ${PYPI_PASSWORD} --verbose 

README.md

@@ -10,6 +10,293 @@
 
   <br>
 
+[![version](https://badge.fury.io/py/supervision.svg)](https://badge.fury.io/py/supervision)
+[![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)
+
+</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! 🤝
+
+## 💻 install
+
+Pip install the supervision package in a
+[**3.11>=Python>=3.8**](https://www.python.org/) environment.
+
+```bash
+pip install supervision[desktop]
+```
+
+Read more about desktop, headless, and local installation in our [guide](https://roboflow.github.io/supervision/).
+
+## 🔥 quickstart
+
+### [detections processing](https://roboflow.github.io/supervision/detection/core/)
+
+```python
+>>> import supervision as sv
+>>> from ultralytics import YOLO
+
+>>> model = YOLO('yolov8s.pt')
+>>> result = model(IMAGE)[0]
+>>> detections = sv.Detections.from_yolov8(result)
+
+>>> len(detections)
+5
+```
+
+<details close>
+<summary>👉 more detections utils</summary>
+  
+- Easily switch inference pipeline between supported object detection/instance segmentation models
+  
+    ```python
+    >>> import supervision as sv
+    >>> from segment_anything import sam_model_registry, SamAutomaticMaskGenerator
+
+    >>> sam = sam_model_registry[MODEL_TYPE](checkpoint=CHECKPOINT_PATH).to(device=DEVICE)
+    >>> mask_generator = SamAutomaticMaskGenerator(sam)
+    >>> sam_result = mask_generator.generate(IMAGE)
+    >>> detections = sv.Detections.from_sam(sam_result=sam_result)
+    ```
+ 
+- [Advanced filtering](https://roboflow.github.io/supervision/quickstart/detections/)
+  
+    ```python
+    >>> detections = detections[detections.class_id == 0]
+    >>> detections = detections[detections.confidence > 0.5]
+    >>> detections = detections[detections.area > 1000]
+    ```
+  
+- Image annotation
+  
+    ```python
+    >>> import supervision as sv
+
+    >>> box_annotator = sv.BoxAnnotator()
+    >>> annotated_frame = box_annotator.annotate(
+    ...     scene=IMAGE,
+    ...     detections=detections
+    ... )
+    ```
+  
+</details>
+
+### [datasets processing](https://roboflow.github.io/supervision/dataset/core/)
+
+```python
+>>> import supervision as sv
+
+>>> dataset = sv.DetectionDataset.from_yolo(
+...     images_directory_path='...',
+...     annotations_directory_path='...',
+...     data_yaml_path='...'
+... )
+
+>>> dataset.classes
+['dog', 'person']
+
+>>> len(dataset)
+1000
+```
+
+<details close>
+<summary>👉 more dataset utils</summary>
+
+- Load object detection/instance segmentation datasets in one of the supported formats
+
+    ```python
+    >>> dataset = sv.DetectionDataset.from_yolo(
+    ...     images_directory_path='...',
+    ...     annotations_directory_path='...',
+    ...     data_yaml_path='...'
+    ... )
+
+    >>> dataset = sv.DetectionDataset.from_pascal_voc(
+    ...     images_directory_path='...',
+    ...     annotations_directory_path='...'
+    ... )
+
+    >>> dataset = sv.DetectionDataset.from_coco(
+    ...     images_directory_path='...',
+    ...     annotations_path='...'
+    ... )
+    ```
+  
+- Loop over dataset entries
+
+    ```python
+    >>> for name, image, labels in dataset:
+    ...     print(labels.xyxy)
+
+    array([[404.      , 719.      , 538.      , 884.5     ],
+           [155.      , 497.      , 404.      , 833.5     ],
+           [ 20.154999, 347.825   , 416.125   , 915.895   ]], dtype=float32)
+    ```
+  
+- Split dataset for training, testing, and validation
+  
+    ```python
+    >>> train_dataset, test_dataset = dataset.split(split_ratio=0.7)
+    >>> test_dataset, valid_dataset = test_dataset.split(split_ratio=0.5)
+  
+    >>> len(train_dataset), len(test_dataset), len(valid_dataset)
+    (700, 150, 150)
+    ```
+  
+- Merge multiple datasets
+
+    ```python
+    >>> ds_1 = sv.DetectionDataset(...)
+    >>> len(ds_1)
+    100
+    >>> ds_1.classes
+    ['dog', 'person']
+  
+    >>> ds_2 = sv.DetectionDataset(...)
+    >>> len(ds_2)
+    200
+    >>> ds_2.classes
+    ['cat']
+  
+    >>> ds_merged = sv.DetectionDataset.merge([ds_1, ds_2])
+    >>> len(ds_merged)
+    300
+    >>> ds_merged.classes
+    ['cat', 'dog', 'person']
+    ```
+  
+- Save object detection/instance segmentation datasets in one of the supported formats
+  
+    ```python
+    >>> dataset.as_yolo(
+    ...     images_directory_path='...',
+    ...     annotations_directory_path='...',
+    ...     data_yaml_path='...'
+    ... )
+
+    >>> dataset.as_pascal_voc(
+    ...     images_directory_path='...',
+    ...     annotations_directory_path='...'
+    ... )
+
+    >>> dataset.as_coco(
+    ...     images_directory_path='...',
+    ...     annotations_path='...'
+    ... )
+    ```
+  
+- Convert labels between supported formats
+  
+    ```python
+    >>> sv.DetectionDataset.from_yolo(
+    ...     images_directory_path='...',
+    ...     annotations_directory_path='...',
+    ...     data_yaml_path='...'
+    ... ).as_pascal_voc(
+    ...     images_directory_path='...',
+    ...     annotations_directory_path='...'
+    ... )
+    ```
+  
+- Load classification datasets in one of the supported formats
+
+    ```python
+    >>> cs = sv.ClassificationDataset.from_folder_structure(
+    ...     root_directory_path='...'
+    ... )
+    ```
+
+- Save classification datasets in one of the supported formats
+
+    ```python
+    >>> cs.as_folder_structure(
+    ...     root_directory_path='...'
+    ... )
+    ```
+
+</details>
+
+### [model evaluation](https://roboflow.github.io/supervision/metrics/detection/)
+
+```python
+>>> import supervision as sv
+
+>>> dataset = sv.DetectionDataset.from_yolo(...)
+
+>>> def callback(image: np.ndarray) -> sv.Detections:
+...     ...
+
+>>> confusion_matrix = sv.ConfusionMatrix.benchmark(
+...     dataset = dataset,
+...     callback = callback
+... )
+
+>>> confusion_matrix.matrix
+array([
+    [0., 0., 0., 0.],
+    [0., 1., 0., 1.],
+    [0., 1., 1., 0.],
+    [1., 1., 0., 0.]
+])
+```
+
+<details close>
+<summary>👉 more metrics</summary>
+
+- Mean average precision (mAP) for object detection tasks.
+
+    ```python
+    >>> import supervision as sv
+    
+    >>> dataset = sv.DetectionDataset.from_yolo(...)
+    
+    >>> def callback(image: np.ndarray) -> sv.Detections:
+    ...     ...
+    
+    >>> mean_average_precision = sv.MeanAveragePrecision.benchmark(
+    ...     dataset = dataset,
+    ...     callback = callback
+    ... )
+    
+    >>> mean_average_precision.map50_95
+    0.433 
+    ```
+  
+</details>
+
+## 🎬 tutorials
+
+<p align="left">
+<a href="https://youtu.be/oEQYStnF2l8" title="Accelerate Image Annotation with SAM and Grounding DINO"><img src="https://github.com/SkalskiP/SkalskiP/assets/26109316/ae1ca38e-40b7-4b35-8582-e8ea5de3806e" alt="Accelerate Image Annotation with SAM and Grounding DINO" width="300px" align="left" /></a>
+<a href="https://youtu.be/oEQYStnF2l8" title="Accelerate Image Annotation with SAM and Grounding DINO"><strong>Accelerate Image Annotation with SAM and Grounding DINO</strong></a>
+<div><strong>Created: 20 Apr 2023</strong> | <strong>Updated: 20 Apr 2023</strong></div>
+<br/> Discover how to speed up your image annotation process using Grounding DINO and Segment Anything Model (SAM). Learn how to convert object detection datasets into instance segmentation datasets, and see the potential of using these models to automatically annotate your datasets for real-time detectors like YOLOv8... </p> 
+
+<br/> 
+
+<p align="left">
+<a href="https://youtu.be/oEQYStnF2l8" title="SAM - Segment Anything Model by Meta AI: Complete Guide"><img src="https://github.com/SkalskiP/SkalskiP/assets/26109316/6913ff11-53c6-4341-8d90-eaff3023c3fd" alt="SAM - Segment Anything Model by Meta AI: Complete Guide" width="300px" align="left" /></a>
+<a href="https://youtu.be/oEQYStnF2l8" title="SAM - Segment Anything Model by Meta AI: Complete Guide"><strong>SAM - Segment Anything Model by Meta AI: Complete Guide</strong></a>
+<div><strong>Created: 11 Apr 2023</strong> | <strong>Updated: 11 Apr 2023</strong></div>
+<br/> Discover the incredible potential of Meta AI's Segment Anything Model (SAM)! We dive into SAM, an efficient and promptable model for image segmentation, which has revolutionized computer vision tasks. With over 1 billion masks on 11M licensed and privacy-respecting images, SAM's zero-shot performance is often competitive with or even superior to prior fully supervised results... </p>
+
+## 📚 documentation
+
+Visit our [documentation](https://roboflow.github.io/supervision) page to learn how supervision can help you build computer vision applications faster and more reliably. 
+
+## 🏆 contribution
+
+We love your input! Please see our [contributing guide](https://github.com/roboflow/supervision/blob/main/CONTRIBUTING.md) to get started. Thank you 🙏 to all our contributors!
+
+<br>
+
+<div align="center">
+
   <div align="center">
       <a href="https://youtube.com/roboflow">
           <img
@@ -54,73 +341,4 @@
       </a>
   </div>
 
-  <br>
-
-[![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)
-
 </div>
-
-## 👋 hello
-
-A set of easy-to-use utils that will come in handy in any computer vision project. **Supervision is still in 
-pre-release stage. 🚧 Keep your eyes open for potential bugs and be aware that at this stage our API is still fluid 
-and may change.**
-
-## 💻 install
-
-Pip install the supervision package in a
-[**3.10>=Python>=3.7**](https://www.python.org/) environment.
-
-```bash
-pip install supervision
-```
-
-<details close>
-<summary>Install from source</summary>
-
-```bash
-# clone repository and navigate to root directory
-git clone https://github.com/roboflow/supervision.git
-cd supervision
-
-# setup python environment and activate it
-python3 -m venv venv
-source venv/bin/activate
-
-# install
-pip install -e ".[dev]"
-```
-
-</details>
-
-## 📖 documentation
-
-Curious how Supervision can help you solve problems on your project? Visit our [documentation](https://roboflow.github.io/supervision) page!
-
-## 🎬 tutorials
-
-🔥 [Subscribe](https://www.youtube.com/@Roboflow) and stay up to date with the latest changes to the Supervision library. 
-
-<p align="center">
-    <a href="https://youtu.be/l_kf9CfZ_8M">
-        <img 
-            width="90%"
-            src="https://user-images.githubusercontent.com/26109316/217950212-311de186-1862-4b4c-a86e-89cafd68b233.jpg" 
-            alt="count-objects-in-zone"
-        >
-    </a>
-</p>
-
-<p align="center">
-    <a href="https://youtu.be/OS5qI9YBkfk">
-        <img
-            width="90%"
-            src="https://user-images.githubusercontent.com/26109316/213702005-ddd568f0-b902-46c2-9af9-e6ff33db23bf.jpg" 
-            alt="count-objects-crossing-line"
-        >
-    </a>
-</p>
-
-## 🏆 contribution
-
-We love your input! Please see our [contributing guide](https://github.com/roboflow/supervision/blob/main/CONTRIBUTING.md) to get started. Thank you 🙏 to all our contributors!

docs/annotation/voc.md

@@ -1,3 +0,0 @@
-## detections_to_voc_xml
-
-:::supervision.annotation.voc.detections_to_voc_xml

docs/changelog.md

@@ -1,27 +1,278 @@
+### 0.13.0 <small>August 8, 2023</small>
+
+- Added [#236](https://github.com/roboflow/supervision/pull/236): support for mean average precision (mAP) for object detection models with [`sv.MeanAveragePrecision`](https://roboflow.github.io/supervision/metrics/detection/#meanaverageprecision). 
+
+```python
+>>> import supervision as sv
+>>> from ultralytics import YOLO
+
+>>> dataset = sv.DetectionDataset.from_yolo(...)
+
+>>> model = YOLO(...)
+>>> def callback(image: np.ndarray) -> sv.Detections:
+...     result = model(image)[0]
+...     return sv.Detections.from_yolov8(result)
+
+>>> mean_average_precision = sv.MeanAveragePrecision.benchmark(
+...     dataset = dataset,
+...     callback = callback
+... )
+
+>>> mean_average_precision.map50_95
+0.433
+```
+
+- Added [#256](https://github.com/roboflow/supervision/pull/256): support for ByteTrack for object tracking with [`sv.ByteTrack`](https://roboflow.github.io/supervision/tracker/core/#bytetrack).
+
+- Added [#222](https://github.com/roboflow/supervision/pull/222): [`sv.Detections.from_ultralytics`](https://roboflow.github.io/supervision/detection/core/#supervision.detection.core.Detections.from_ultralytics) to enable seamless integration with [Ultralytics](https://github.com/ultralytics/ultralytics) framework. This will enable you to use `supervision` with all [models](https://docs.ultralytics.com/models/) that Ultralytics supports.
+
+!!! warning
+
+    [`sv.Detections.from_yolov8`](https://roboflow.github.io/supervision/detection/core/#supervision.detection.core.Detections.from_yolov8) is now deprecated and will be removed with `supervision-0.15.0` release.
+
+- Added [#191](https://github.com/roboflow/supervision/pull/191): [`sv.Detections.from_paddledet`](https://roboflow.github.io/supervision/detection/core/#supervision.detection.core.Detections.from_paddledet) to enable seamless integration with [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection) framework.
+
+- Added [#245](https://github.com/roboflow/supervision/pull/245): support for loading PASCAL VOC segmentation datasets with [`sv.DetectionDataset.`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.DetectionDataset.from_pascal_voc).
+
+### 0.12.0 <small>July 24, 2023</small>
+
+!!! warning
+
+    With the `supervision-0.12.0` release, we are terminating official support for Python 3.7.
+
+- Added [#177](https://github.com/roboflow/supervision/pull/177): initial support for object detection model benchmarking with [`sv.ConfusionMatrix`](https://roboflow.github.io/supervision/metrics/detection/#confusionmatrix).
+
+```python
+>>> import supervision as sv
+>>> from ultralytics import YOLO
+
+>>> dataset = sv.DetectionDataset.from_yolo(...)
+
+>>> model = YOLO(...)
+>>> def callback(image: np.ndarray) -> sv.Detections:
+...     result = model(image)[0]
+...     return sv.Detections.from_yolov8(result)
+
+>>> confusion_matrix = sv.ConfusionMatrix.benchmark(
+...     dataset = dataset,
+...     callback = callback
+... )
+
+>>> confusion_matrix.matrix
+array([
+    [0., 0., 0., 0.],
+    [0., 1., 0., 1.],
+    [0., 1., 1., 0.],
+    [1., 1., 0., 0.]
+])
+```
+
+- Added [#173](https://github.com/roboflow/supervision/pull/173): [`Detections.from_mmdetection`](https://roboflow.github.io/supervision/detection/core/#supervision.detection.core.Detections.from_mmdetection) to enable seamless integration with [MMDetection](https://github.com/open-mmlab/mmdetection) framework.
+
+- Added [#130](https://github.com/roboflow/supervision/issues/130): ability to [install](https://roboflow.github.io/supervision/) package in `headless` or `desktop` mode.
+
+- Changed [#180](https://github.com/roboflow/supervision/pull/180): packing method from `setup.py` to `pyproject.toml`.
+
+- Fixed [#188](https://github.com/roboflow/supervision/issues/188): [`sv.DetectionDataset.from_cooc`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.DetectionDataset.from_coco) can't be loaded when there are images without annotations.
+
+- Fixed [#226](https://github.com/roboflow/supervision/issues/226): [`sv.DetectionDataset.from_yolo`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.DetectionDataset.from_yolo) can't load background instances.
+
+### 0.11.1 <small>June 29, 2023</small>
+
+- Fix [#165](https://github.com/roboflow/supervision/pull/165): [`as_folder_structure`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.ClassificationDataset.as_folder_structure) fails to save [`sv.ClassificationDataset`](https://roboflow.github.io/supervision/dataset/core/#classificationdataset) when it is result of inference.
+
+### 0.11.0 <small>June 28, 2023</small>
+
+- Added [#150](https://github.com/roboflow/supervision/pull/150): ability to load and save [`sv.DetectionDataset`](https://roboflow.github.io/supervision/dataset/core/#detectiondataset) in COCO format using [`as_coco`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.DetectionDataset.as_coco) and [`from_coco`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.DetectionDataset.from_coco) methods. 
+
+```python
+>>> import supervision as sv
+
+>>> ds = sv.DetectionDataset.from_coco(
+...     images_directory_path='...',
+...     annotations_path='...'
+... )
+
+>>> ds.as_coco(
+...     images_directory_path='...',
+...     annotations_path='...'
+... )
+```
+
+- Added [#158](https://github.com/roboflow/supervision/pull/158): ability to marge multiple [`sv.DetectionDataset`](https://roboflow.github.io/supervision/dataset/core/#detectiondataset) together using [`merge`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.DetectionDataset.merge) method. 
+
+```python
+>>> import supervision as sv
+
+>>> ds_1 = sv.DetectionDataset(...)
+>>> len(ds_1)
+100
+>>> ds_1.classes
+['dog', 'person']
+
+>>> ds_2 = sv.DetectionDataset(...)
+>>> len(ds_2)
+200
+>>> ds_2.classes
+['cat']
+
+>>> ds_merged = sv.DetectionDataset.merge([ds_1, ds_2])
+>>> len(ds_merged)
+300
+>>> ds_merged.classes
+['cat', 'dog', 'person']
+```
+
+- Added [#162](https://github.com/roboflow/supervision/pull/162): additional `start` and `end` arguments to [`sv.get_video_frames_generator`](https://roboflow.github.io/supervision/utils/video/#get_video_frames_generator) allowing to generate frames only for a selected part of the video.
+
+- Fix [#157](https://github.com/roboflow/supervision/pull/157): incorrect loading of YOLO dataset class names from `data.yaml`.
+
+### 0.10.0 <small>June 14, 2023</small>
+
+- Added [#125](https://github.com/roboflow/supervision/pull/125): ability to load and save [`sv.ClassificationDataset`](https://roboflow.github.io/supervision/dataset/core/#classificationdataset) in a folder structure format.
+
+```python
+>>> import supervision as sv
+
+>>> cs = sv.ClassificationDataset.from_folder_structure(
+...     root_directory_path='...'
+... )
+
+>>> cs.as_folder_structure(
+...     root_directory_path='...'
+... )
+```
+
+- Added [#125](https://github.com/roboflow/supervision/pull/125): support for [`sv.ClassificationDataset.split`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.ClassificationDataset.split) allowing to divide `sv.ClassificationDataset` into two parts.
+
+- Added [#110](https://github.com/roboflow/supervision/pull/110): ability to extract masks from Roboflow API results using [`sv.Detections.from_roboflow`](https://roboflow.github.io/supervision/detection/core/#supervision.detection.core.Detections.from_roboflow). 
+
+- Added [commit hash](https://github.com/roboflow/supervision/commit/d000292eb2f2342544e0947b65528082e60fb8d6): Supervision Quickstart [notebook](https://colab.research.google.com/github/roboflow/supervision/blob/main/demo.ipynb) where you can learn more about Detection, Dataset and Video APIs.
+
+- Changed [#135](https://github.com/roboflow/supervision/pull/135): `sv.get_video_frames_generator` documentation to better describe actual behavior.  
+
+### 0.9.0 <small>June 7, 2023</small>
+
+- Added [#118](https://github.com/roboflow/supervision/pull/118): ability to select [`sv.Detections`](https://roboflow.github.io/supervision/detection/core/#supervision.detection.core.Detections.__getitem__) by index, list of indexes or slice. Here is an example illustrating the new selection methods.
+
+```python
+>>> import supervision as sv
+
+>>> detections = sv.Detections(...)
+>>> len(detections[0])
+1
+>>> len(detections[[0, 1]])
+2
+>>> len(detections[0:2])
+2
+```
+
+- Added [#101](https://github.com/roboflow/supervision/pull/101): ability to extract masks from YOLOv8 result using [`sv.Detections.from_yolov8`](https://roboflow.github.io/supervision/detection/core/#supervision.detection.core.Detections.from_yolov8). Here is an example illustrating how to extract boolean masks from the result of the YOLOv8 model inference.
+ 
+- Added [#122](https://github.com/roboflow/supervision/pull/122): ability to crop image using [`sv.crop`](https://roboflow.github.io/supervision/utils/image/#crop). Here is an example showing how to get a separate crop for each detection in `sv.Detections`.
+
+- Added [#120](https://github.com/roboflow/supervision/pull/120): ability to conveniently save multiple images into directory using [`sv.ImageSink`](https://roboflow.github.io/supervision/utils/image/#imagesink). Here is an example showing how to save every tenth video frame as a separate image.
+
+```python
+>>> import supervision as sv
+
+>>> with sv.ImageSink(target_dir_path='target/directory/path') as sink:
+...     for image in sv.get_video_frames_generator(source_path='source_video.mp4', stride=10):
+...         sink.save_image(image=image)
+```
+
+- Fixed [#106](https://github.com/roboflow/supervision/issues/106): inconvenient handling of [`sv.PolygonZone`](https://roboflow.github.io/supervision/detection/tools/polygon_zone/#polygonzone) coordinates. Now `sv.PolygonZone` accepts coordinates in the form of `[[x1, y1], [x2, y2], ...]` that can be both integers and floats.
+
+### 0.8.0 <small>May 17, 2023</small>
+
+- Added [#100](https://github.com/roboflow/supervision/pull/100): support for dataset inheritance. The current `Dataset` got renamed to `DetectionDataset`. Now [`DetectionDataset`](https://roboflow.github.io/supervision/dataset/core/#detectiondataset) inherits from `BaseDataset`. This change was made to enforce the future consistency of APIs of different types of computer vision datasets.
+- Added [#100](https://github.com/roboflow/supervision/pull/100): ability to save datasets in YOLO format using [`DetectionDataset.as_yolo`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.DetectionDataset.as_yolo). 
+
+```python
+>>> import roboflow
+>>> from roboflow import Roboflow
+>>> import supervision as sv
+
+>>> roboflow.login()
+
+>>> rf = Roboflow()
+
+>>> project = rf.workspace(WORKSPACE_ID).project(PROJECT_ID)
+>>> dataset = project.version(PROJECT_VERSION).download("yolov5")
+
+>>> ds = sv.DetectionDataset.from_yolo(
+...     images_directory_path=f"{dataset.location}/train/images",
+...     annotations_directory_path=f"{dataset.location}/train/labels",
+...     data_yaml_path=f"{dataset.location}/data.yaml"
+... )
+
+>>> ds.classes
+['dog', 'person']
+```
+
+- Added [#102](https://github.com/roboflow/supervision/pull/103): support for [`DetectionDataset.split`](https://roboflow.github.io/supervision/dataset/core/#supervision.dataset.core.DetectionDataset.split) allowing to divide `DetectionDataset` into two parts. 
+
+```python
+>>> import supervision as sv
+
+>>> ds = sv.DetectionDataset(...)
+>>> train_ds, test_ds = ds.split(split_ratio=0.7, random_state=42, shuffle=True)
+
+>>> len(train_ds), len(test_ds)
+(700, 300)
+```
+
+- Changed [#100](https://github.com/roboflow/supervision/pull/100): default value of `approximation_percentage` parameter from `0.75` to `0.0` in `DetectionDataset.as_yolo` and `DetectionDataset.as_pascal_voc`.
+
+### 0.7.0 <small>May 11, 2023</small>
+
+- Added [#91](https://github.com/roboflow/supervision/pull/91): `Detections.from_yolo_nas` to enable seamless integration with [YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md) model.
+- Added [#86](https://github.com/roboflow/supervision/pull/86): ability to load datasets in YOLO format using `Dataset.from_yolo`. 
+- Added [#84](https://github.com/roboflow/supervision/pull/84): `Detections.merge` to merge multiple `Detections` objects together.
+- Fixed [#81](https://github.com/roboflow/supervision/pull/81): `LineZoneAnnotator.annotate` does not return annotated frame.
+- Changed [#44](https://github.com/roboflow/supervision/pull/44): `LineZoneAnnotator.annotate` to allow for custom text for the in and out tags.
+
+### 0.6.0 <small>April 19, 2023</small>
+
+- Added [#71](https://github.com/roboflow/supervision/pull/71): initial `Dataset` support and ability to save `Detections` in Pascal VOC XML format. 
+- Added [#71](https://github.com/roboflow/supervision/pull/71): new `mask_to_polygons`, `filter_polygons_by_area`, `polygon_to_xyxy` and `approximate_polygon` utilities.
+- Added [#72](https://github.com/roboflow/supervision/pull/72): ability to load Pascal VOC XML **object detections** dataset as `Dataset`.
+- Changed [#70](https://github.com/roboflow/supervision/pull/70): order of `Detections` attributes to make it consistent with order of objects in `__iter__` tuple.
+- Changed [#71](https://github.com/roboflow/supervision/pull/71): `generate_2d_mask` to `polygon_to_mask`.
+
+### 0.5.2 <small>April 13, 2023</small>
+
+- Fixed [#63](https://github.com/roboflow/supervision/pull/63): `LineZone.trigger` function expects 4 values instead of 5.
+
+### 0.5.1 <small>April 12, 2023</small>
+
+- Fixed `Detections.__getitem__` method did not return mask for selected item.
+- Fixed `Detections.area` crashed for mask detections.
+
 ### 0.5.0 <small>April 10, 2023</small>
 
-- Added [[#58](https://github.com/roboflow/supervision/pull/58)]: `Detections.mask` to enable segmentation support.
-- Added [[#58](https://github.com/roboflow/supervision/pull/58)]: `MaskAnnotator` to allow easy `Detections.mask` annotation.
-- Added [[#58](https://github.com/roboflow/supervision/pull/58)]: `Detections.from_sam` to enable native Segment Anything Model (SAM) support.
-- Changed [[#58](https://github.com/roboflow/supervision/pull/58)]: `Detections.area` behaviour to work not only with boxes but also with masks.
+- Added [#58](https://github.com/roboflow/supervision/pull/58): `Detections.mask` to enable segmentation support.
+- Added [#58](https://github.com/roboflow/supervision/pull/58): `MaskAnnotator` to allow easy `Detections.mask` annotation.
+- Added [#58](https://github.com/roboflow/supervision/pull/58): `Detections.from_sam` to enable native Segment Anything Model (SAM) support.
+- Changed [#58](https://github.com/roboflow/supervision/pull/58): `Detections.area` behaviour to work not only with boxes but also with masks.
 
 ### 0.4.0 <small>April 5, 2023</small> 
 
-- Added [[#46](https://github.com/roboflow/supervision/discussions/48)]: `Detections.empty` to allow easy creation of empty `Detections` objects.
-- Added [[#56](https://github.com/roboflow/supervision/pull/56)]: `Detections.from_roboflow` to allow easy creation of `Detections` objects from Roboflow API inference results.
-- Added [[#56](https://github.com/roboflow/supervision/pull/56)]: `plot_images_grid` to allow easy plotting of multiple images on single plot.
-- Added [[#56](https://github.com/roboflow/supervision/pull/56)]: initial support for Pascal VOC XML format with `detections_to_voc_xml` method.
-- Changed [[#56](https://github.com/roboflow/supervision/pull/56)]: `show_frame_in_notebook` refactored and renamed to `plot_image`.
+- Added [#46](https://github.com/roboflow/supervision/discussions/48): `Detections.empty` to allow easy creation of empty `Detections` objects.
+- Added [#56](https://github.com/roboflow/supervision/pull/56): `Detections.from_roboflow` to allow easy creation of `Detections` objects from Roboflow API inference results.
+- Added [#56](https://github.com/roboflow/supervision/pull/56): `plot_images_grid` to allow easy plotting of multiple images on single plot.
+- Added [#56](https://github.com/roboflow/supervision/pull/56): initial support for Pascal VOC XML format with `detections_to_voc_xml` method.
+- Changed [#56](https://github.com/roboflow/supervision/pull/56): `show_frame_in_notebook` refactored and renamed to `plot_image`.
 
 ### 0.3.2 <small>March 23, 2023</small> 
 
-- Changed [[#50](https://github.com/roboflow/supervision/issues/50)]: Allow `Detections.class_id` to be `None`. 
+- Changed [#50](https://github.com/roboflow/supervision/issues/50): Allow `Detections.class_id` to be `None`. 
 
 ### 0.3.1 <small>March 6, 2023</small> 
 
-- Fixed [[#41](https://github.com/roboflow/supervision/issues/41)]: `PolygonZone` throws an exception when the object touches the bottom edge of the image.
-- Fixed [[#42](https://github.com/roboflow/supervision/issues/42)]: `Detections.wth_nms` method throws an exception when `Detections` is empty.
-- Changed [[#36](https://github.com/roboflow/supervision/pull/36)]: `Detections.wth_nms` support class agnostic and non-class agnostic case.
+- Fixed [#41](https://github.com/roboflow/supervision/issues/41): `PolygonZone` throws an exception when the object touches the bottom edge of the image.
+- Fixed [#42](https://github.com/roboflow/supervision/issues/42): `Detections.wth_nms` method throws an exception when `Detections` is empty.
+- Changed [#36](https://github.com/roboflow/supervision/pull/36): `Detections.wth_nms` support class agnostic and non-class agnostic case.
 
 ### 0.3.0 <small>March 6, 2023</small> 
 

docs/classification/core.md

@@ -0,0 +1,3 @@
+## Classifications
+
+:::supervision.classification.core.Classifications

docs/dataset/core.md

@@ -0,0 +1,12 @@
+!!! warning
+
+    Dataset API is still fluid and may change. If you use Dataset API in your project until further notice, freeze the 
+    `supervision` version in your `requirements.txt` or `setup.py`.
+
+## DetectionDataset
+
+:::supervision.dataset.core.DetectionDataset
+
+## ClassificationDataset
+
+:::supervision.dataset.core.ClassificationDataset

docs/detection/utils.md

@@ -1,7 +1,3 @@
-## generate_2d_mask
-
-:::supervision.detection.utils.generate_2d_mask
-
 ## box_iou_batch
 
 :::supervision.detection.utils.box_iou_batch
@@ -10,6 +6,22 @@
 
 :::supervision.detection.utils.non_max_suppression
 
+## polygon_to_mask
+
+:::supervision.detection.utils.polygon_to_mask
+
 ## mask_to_xyxy
 
-:::supervision.detection.utils.mask_to_xyxy
+:::supervision.detection.utils.mask_to_xyxy
+
+## mask_to_polygons
+
+:::supervision.detection.utils.mask_to_polygons
+
+## polygon_to_xyxy
+
+:::supervision.detection.utils.polygon_to_xyxy
+
+## filter_polygons_by_area
+
+:::supervision.detection.utils.filter_polygons_by_area

docs/index.md

@@ -9,29 +9,65 @@
   </p>
 </div>
 
-## 👋 Welcome
+## 👋 Hello
 
-Supervision is a set of easy-to-use utilities that will come in handy in any computer vision project. 
+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** is still in 
-pre-release stage 🚧 Keep your eyes open for potential bugs and be aware that at this stage our API is still fluid and may change.
-
-## 💻 How to Install
+## 💻 Install
 
 You can install `supervision` with pip in a 
-[**3.10>=Python>=3.7**](https://www.python.org/) environment.
+[**3.11>=Python>=3.8**](https://www.python.org/) environment.
 
-!!! example "Pip install method (recommended)"
+!!! example "pip install (recommended)"
 
-    ```bash
-    pip install supervision
-    ```
+    === "headless"
+        The headless installation of `supervision` is designed for environments where graphical user interfaces (GUI) are not needed, making it more lightweight and suitable for server-side applications. 
 
-!!! example "Git clone method (for development)"
+        ```bash
+        pip install supervision
+        ```
+    
+    === "desktop"
+        If you require the full version of `supervision` with GUI support you can install the desktop version. This version includes the GUI components of OpenCV, allowing you to display images and videos on the screen.
 
-    ```bash
-    git https://github.com/roboflow/supervision.git
-    cd supervision
-    pip install -e '.[dev]'
-    ```
-    See contributing section to know more about contributing to the project
+        ```bash
+        pip install supervision[desktop]
+        ```
+
+!!! example "git clone (for development)"
+
+    === "virtualenv"
+
+        ```bash
+        # clone repository and navigate to root directory
+        git clone https://github.com/roboflow/supervision.git
+        cd supervision
+        
+        # setup python environment and activate it
+        python3 -m venv venv
+        source venv/bin/activate
+        
+        # headless install
+        pip install -e "."
+        
+        # desktop install
+        pip install -e ".[desktop]"
+        ```
+    
+    === "poetry"
+
+        ```bash
+        # clone repository and navigate to root directory
+        git clone https://github.com/roboflow/supervision.git
+        cd supervision
+ 
+        # setup python environment and activate it
+        poetry env use python 3.10
+        poetry shell
+
+        # headless install
+        poetry install
+        
+        # desktop install
+        poetry install --extras "desktop"
+        ```

docs/metrics/detection.md

@@ -0,0 +1,12 @@
+!!! 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`.
+
+## ConfusionMatrix
+
+:::supervision.metrics.detection.ConfusionMatrix
+
+## MeanAveragePrecision
+
+:::supervision.metrics.detection.MeanAveragePrecision

docs/notebook/utils.md

@@ -1,7 +0,0 @@
-## plot_image
-
-:::supervision.notebook.utils.plot_image
-
-## plot_images_grid
-
-:::supervision.notebook.utils.plot_images_grid

docs/quickstart/detections.md

@@ -0,0 +1,312 @@
+## advanced filtering
+
+
+The advanced filtering capabilities of the `Detections` class offer users a versatile and efficient way to narrow down 
+and refine object detections. This section outlines various filtering methods, including filtering by specific class 
+or a set of classes, confidence, object area, bounding box area, relative area, box dimensions, and designated zones. 
+Each method is demonstrated with concise code examples to provide users with a clear understanding of how to implement 
+the filters in their applications.
+
+### by specific class
+
+Allows you to select detections that belong only to one selected class.
+
+=== "After"
+
+    ```python
+    import supervision as sv
+
+    detections = sv.Detections(...)
+    detections = detections[detections.class_id == 0]
+    ```
+
+    <div class="result" markdown>
+
+    ![by-specific-class](https://media.roboflow.com/open-source/supervision/supervision-detection-by-specific-class.png){ align=center width="800" }
+
+    </div>
+
+=== "Before"
+
+    ```python
+    import supervision as sv
+
+    detections = sv.Detections(...)
+    detections = detections[detections.class_id == 0]
+    ```
+
+    <div class="result" markdown>
+
+    ![original](https://media.roboflow.com/open-source/supervision/supervision-detection-original.png){ align=center width="800" }
+
+    </div>
+
+
+### by set of classes
+
+Allows you to select detections that belong only to selected set of classes. 
+
+=== "After"
+
+    ```python
+    import numpy as np
+    import supervision as sv
+    
+    selected_classes = [0, 2, 3]
+    detections = sv.Detections(...)
+    detections = detections[np.isin(detections.class_id, selected_classes)]
+    ```
+
+    <div class="result" markdown>
+
+    ![by-set-of-classes](https://media.roboflow.com/open-source/supervision/supervision-detection-by-set-of-classes.png){ align=center width="800" }
+
+    </div>
+
+=== "Before"
+
+    ```python
+    import numpy as np
+    import supervision as sv
+    
+    class_id = [0, 2, 3]
+    detections = sv.Detections(...)
+    detections = detections[np.isin(detections.class_id, class_id)]
+    ```
+
+    <div class="result" markdown>
+
+    ![original](https://media.roboflow.com/open-source/supervision/supervision-detection-original.png){ align=center width="800" }
+
+    </div>
+
+### by confidence
+
+Allows you to select detections with specific confidence value, for example higher than selected threshold.
+
+=== "After"
+
+    ```python
+    import supervision as sv
+
+    detections = sv.Detections(...)
+    detections = detections[detections.confidence > 0.5]
+    ```
+
+    <div class="result" markdown>
+
+    ![by-set-of-classes](https://media.roboflow.com/open-source/supervision/supervision-detection-by-confidence.png){ align=center width="800" }
+
+    </div>
+
+=== "Before"
+
+    ```python
+    import supervision as sv
+
+    detections = sv.Detections(...)
+    detections = detections[detections.confidence > 0.5]
+    ```
+
+    <div class="result" markdown>
+
+    ![original](https://media.roboflow.com/open-source/supervision/supervision-detection-original.png){ align=center width="800" }
+
+    </div>
+
+### by area
+
+Allows you to select detections based on their size. We define the area as the number of pixels occupied by the 
+detection in the image. In the example below, we have sifted out the detections that are too small. 
+
+=== "After"
+
+    ```python
+    import supervision as sv
+
+    detections = sv.Detections(...)
+    detections = detections[detections.area > 1000]
+    ```
+
+    <div class="result" markdown>
+
+    ![by-area](https://media.roboflow.com/open-source/supervision/supervision-detection-by-area.png){ align=center width="800" }
+
+    </div>
+
+=== "Before"
+
+    ```python
+    import supervision as sv
+
+    detections = sv.Detections(...)
+    detections = detections[detections.area > 1000]
+    ```
+
+    <div class="result" markdown>
+
+    ![original](https://media.roboflow.com/open-source/supervision/supervision-detection-original.png){ align=center width="800" }
+
+    </div>
+
+### by relative area
+
+Allows you to select detections based on their size in relation to the size of whole image. Sometimes the concept of 
+detection size changes depending on the image. Detection occupying 10000 square px can be large on a 1280x720 image 
+but small on a 3840x2160 image. In such cases, we can filter out detections based on the percentage of the image area 
+occupied by them. In the example below, we remove too large detections.
+
+=== "After"
+
+    ```python
+    import supervision as sv
+
+    image = ...
+    height, width, channels = image.shape
+    image_area = height * width
+    
+    detections = sv.Detections(...)
+    detections = detections[(detections.area / image_area) < 0.8]
+    ```
+
+    <div class="result" markdown>
+
+    ![by-relative-area](https://media.roboflow.com/open-source/supervision/supervision-detection-by-relative-area.png?updatedAt=1683207183434){ align=center width="800" }
+
+    </div>
+
+=== "Before"
+
+    ```python
+    import supervision as sv
+
+    image = ...
+    height, width, channels = image.shape
+    image_area = height * width
+    
+    detections = sv.Detections(...)
+    detections = detections[(detections.area / image_area) < 0.8]
+    ```
+
+    <div class="result" markdown>
+
+    ![original](https://media.roboflow.com/open-source/supervision/supervision-detection-original.png){ align=center width="800" }
+
+    </div>
+
+### by box dimensions
+
+Allows you to select detections based on their dimensions. The size of the bounding box, as well as its coordinates, 
+can be criteria for rejecting detection. Implementing such filtering requires a bit of custom code but is relatively 
+simple and fast.
+
+=== "After"
+
+    ```python
+    import supervision as sv
+
+    detections = sv.Detections(...)
+    w = detections.xyxy[:, 2] - detections.xyxy[:, 0]
+    h = detections.xyxy[:, 3] - detections.xyxy[:, 1]
+    detections = detections[(w > 200) & (h > 200)]
+    ```
+
+    <div class="result" markdown>
+
+    ![by-box-dimensions](https://media.roboflow.com/open-source/supervision/supervision-detection-by-box-dimensions.png){ align=center width="800" }
+
+    </div>
+
+=== "Before"
+
+    ```python
+    import supervision as sv
+
+    detections = sv.Detections(...)
+    w = detections.xyxy[:, 2] - detections.xyxy[:, 0]
+    h = detections.xyxy[:, 3] - detections.xyxy[:, 1]
+    detections = detections[(w > 200) & (h > 200)]
+    ```
+
+    <div class="result" markdown>
+
+    ![original](https://media.roboflow.com/open-source/supervision/supervision-detection-original.png){ align=center width="800" }
+
+    </div>
+
+### by `PolygonZone`
+
+Allows you to use `Detections` in combination with `PolygonZone` to weed out bounding boxes that are in and out of the 
+zone. In the example below you can see how to filter out all detections located in the lower part of the image.
+
+=== "After"
+
+    ```python
+    import supervision as sv
+
+    zone = sv.PolygonZone(...)
+    detections = sv.Detections(...)
+    mask = zone.trigger(detections=detections)
+    detections = detections[mask]
+    ```
+
+    <div class="result" markdown>
+
+    ![by-polygon-zone](https://media.roboflow.com/open-source/supervision/supervision-detection-by-polygon-zone.png?updatedAt=1683211380445){ align=center width="800" }
+
+    </div>
+
+=== "Before"
+
+    ```python
+    import supervision as sv
+
+    zone = sv.PolygonZone(...)
+    detections = sv.Detections(...)
+    mask = zone.trigger(detections=detections)
+    detections = detections[mask]
+    ```
+
+    <div class="result" markdown>
+
+    ![original](https://media.roboflow.com/open-source/supervision/supervision-detection-original.png){ align=center width="800" }
+
+    </div>
+
+### by mixed conditions
+
+`Detections`' greatest strength, however, is that you can build arbitrarily complex logical conditions by simply combining separate conditions using `&` or `|`.
+
+=== "After"
+
+    ```python
+    import supervision as sv
+
+    zone = sv.PolygonZone(...)
+    detections = sv.Detections(...)
+    mask = zone.trigger(detections=detections)
+    detections = detections[(detections.confidence > 0.7) & mask]
+    ```
+
+    <div class="result" markdown>
+
+    ![by-mixed-conditions](https://media.roboflow.com/open-source/supervision/supervision-detection-by-mixed-conditions.png){ align=center width="800" }
+
+    </div>
+
+=== "Before"
+
+    ```python
+    import supervision as sv
+
+    zone = sv.PolygonZone(...)
+    detections = sv.Detections(...)
+    mask = zone.trigger(detections=detections)
+    detections = detections[mask]
+    ```
+
+    <div class="result" markdown>
+
+    ![original](https://media.roboflow.com/open-source/supervision/supervision-detection-original.png){ align=center width="800" }
+
+    </div>

docs/stylesheets/extra.css

@@ -0,0 +1,4 @@
+:root {
+  --md-primary-fg-color:        #8315F9;
+  --md-accent-fg-color:         #00FFCE;
+}

docs/tracker/core.md

@@ -0,0 +1,4 @@
+## ByteTrack
+
+:::supervision.tracker.byte_tracker.core.ByteTrack
+

docs/utils/file.md

@@ -0,0 +1,3 @@
+## list_files_with_extensions
+
+:::supervision.utils.file.list_files_with_extensions

docs/utils/image.md

@@ -0,0 +1,7 @@
+## ImageSink
+
+:::supervision.utils.image.ImageSink
+
+## crop
+
+:::supervision.utils.image.crop

docs/utils/notebook.md

@@ -0,0 +1,7 @@
+## plot_image
+
+:::supervision.utils.notebook.plot_image
+
+## plot_images_grid
+
+:::supervision.utils.notebook.plot_images_grid

docs/utils/video.md

@@ -0,0 +1,15 @@
+## VideoInfo
+
+:::supervision.utils.video.VideoInfo
+
+## VideoSink
+
+:::supervision.utils.video.VideoSink
+
+## get_video_frames_generator
+
+:::supervision.utils.video.get_video_frames_generator
+
+## process_video
+
+:::supervision.utils.video.process_video

docs/video.md

@@ -1,15 +0,0 @@
-## VideoInfo
-
-:::supervision.video.VideoInfo
-
-## VideoSink
-
-:::supervision.video.VideoSink
-
-## get_video_frames_generator
-
-:::supervision.video.get_video_frames_generator
-
-## process_video
-
-:::supervision.video.process_video

mkdocs.yml

@@ -21,22 +21,35 @@ extra:
     provider: google
     property: G-P7ZG0Y19G5
 
+extra_css:
+  - stylesheets/extra.css
+
 nav:
   - Home: index.md
+  - Quickstart:
+    - Detections: quickstart/detections.md
   - API reference:
-    - Video: video.md
-    - Detection:
+    - Classifications:
+        - Core: classification/core.md
+    - Detections:
         - Core: detection/core.md
         - Annotate: detection/annotate.md
         - Utils: detection/utils.md
         - Tools:
             - Polygon Zone: detection/tools/polygon_zone.md
+    - Trackers:
+        - Core: tracker/core.md
+    - Dataset:
+        - Core: dataset/core.md
+    - Metrics:
+      - Detection Models: metrics/detection.md
     - Draw:
         - Utils: draw/utils.md
-    - Annotations:
-        - Pascal VOC XML: annotation/voc.md
-    - Notebook:
-        - Utils: notebook/utils.md
+    - Utils:
+        - Video: utils/video.md
+        - Image: utils/image.md
+        - Notebook: utils/notebook.md
+        - File: utils/file.md
   - Changelog: changelog.md
 
 theme:
@@ -44,8 +57,19 @@ theme:
   logo: https://media.roboflow.com/open-source/supervision/supervision-lenny.png
   favicon: https://media.roboflow.com/open-source/supervision/supervision-lenny.png
   palette:
-    primary: 'deep purple'
-    accent: 'teal'
+    # Palette for light mode
+    - scheme: default
+      primary: 'custom'
+      toggle:
+        icon: material/brightness-7
+        name: Switch to dark mode
+
+    # Palette toggle for dark mode
+    - scheme: slate
+      primary: 'custom'
+      toggle:
+        icon: material/brightness-4
+        name: Switch to light mode
   font:
     text: Roboto
     code: Roboto Mono
@@ -58,5 +82,9 @@ markdown_extensions:
   - admonition
   - pymdownx.details
   - pymdownx.superfences
+  - attr_list
+  - md_in_html
+  - pymdownx.tabbed:
+      alternate_style: true
   - toc:
       permalink: true

pyproject.toml

@@ -0,0 +1,77 @@
+[tool.poetry]
+name = "supervision"
+version = "0.13.0"
+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>"]
+readme = "README.md"
+license = "MIT"
+packages = [{include = "supervision"}]
+homepage = "https://github.com/roboflow/supervision"
+repository = "https://github.com/roboflow/supervision"
+documentation = "https://github.com/roboflow/supervision/blob/main/README.md"
+keywords = ["machine-learning", "deep-learning", "vision", "ML", "DL", "AI", "YOLOv5", "YOLOv8", "Roboflow"]
+
+classifiers=[
+    'Intended Audience :: Developers',
+    'Intended Audience :: Science/Research',
+    'License :: OSI Approved :: MIT License',
+    'Programming Language :: Python :: 3',
+    'Programming Language :: Python :: 3.8',
+    'Programming Language :: Python :: 3.9',
+    'Programming Language :: Python :: 3.10',
+    'Programming Language :: Python :: 3.11',
+    'Programming Language :: Python :: 3 :: Only',
+    'Topic :: Software Development',
+    'Topic :: Scientific/Engineering',
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Topic :: Scientific/Engineering :: Image Recognition",
+    'Typing :: Typed',
+    'Operating System :: Microsoft :: Windows',
+    'Operating System :: POSIX',
+    'Operating System :: Unix',
+    'Operating System :: MacOS'
+]
+
+[tool.poetry.dependencies]
+python = ">=3.8,<3.12.0"
+numpy = "^1.20.0"
+matplotlib = "^3.7.1"
+pyyaml = "^6.0"
+pillow = "^9.4.0"
+opencv-python = { version = "^4.8.0.74", optional = true }
+opencv-python-headless = "^4.8.0.74"
+scipy = "^1.9.0"
+
+
+[tool.poetry.extras]
+desktop = ["opencv-python"]
+
+
+[tool.poetry.group.dev.dependencies]
+twine = "^4.0.2"
+pytest = "^7.2.2"
+wheel = "^0.40.0"
+notebook = "^6.5.3"
+build = "^0.10.0"
+ruff = "^0.0.280"
+isort = "^5.12.0"
+black = "^23.7.0"
+mypy = "^1.4.1"
+pre-commit = "^3.3.3"
+flake8 = { version = "*", python = ">=3.8.1,<3.12.0" }
+
+
+[tool.poetry.group.docs.dependencies]
+mkdocs-material = "^9.1.4"
+mkdocstrings = {extras = ["python"], version = "^0.20.0"}
+
+[tool.setuptools]
+include-package-data = false
+
+[tool.setuptools.packages.find]
+exclude = ["docs*", "test*"]
+
+[build-system]
+requires = ["poetry-core"]
+build-backend = "poetry.core.masonry.api"

setup.py

@@ -1,67 +0,0 @@
-import setuptools
-from setuptools import find_packages
-import re
-from pathlib import Path
-
-FILE = Path(__file__).resolve()
-PARENT = FILE.parent  # root directory
-README = (PARENT / "README.md").read_text(encoding="utf-8")
-
-
-def get_version():
-    file = PARENT / 'supervision/__init__.py'
-    return re.search(r'^__version__ = [\'"]([^\'"]*)[\'"]', file.read_text(encoding="utf-8"), re.M)[1]
-    
-
-setuptools.setup(
-    name='supervision',
-    version=get_version(),
-    author='Piotr Skalski',
-    author_email='piotr.skalski92@gmail.com',
-    license='MIT',
-    description='A set of easy-to-use utils that will come in handy in any Computer Vision project',
-    long_description=README,
-    long_description_content_type='text/markdown',
-    url='https://github.com/roboflow/supervision',
-    install_requires=[
-        'numpy>=1.20.0',
-        'opencv-python',
-        'matplotlib'
-    ],
-    packages=find_packages(exclude=("tests",)),
-    extras_require={
-        'dev': [
-            'flake8',
-            'black==22.3.0',
-            'isort',
-            'twine',
-            'pytest',
-            'wheel',
-            'notebook',
-            'mkdocs-material',
-            'mkdocstrings[python]'
-        ],
-    },
-    classifiers=[
-        'Intended Audience :: Developers',
-        'Intended Audience :: Science/Research',
-        'License :: OSI Approved :: BSD License',
-        'Programming Language :: Python :: 3',
-        'Programming Language :: Python :: 3.7',
-        'Programming Language :: Python :: 3.8',
-        'Programming Language :: Python :: 3.9',
-        'Programming Language :: Python :: 3.10',
-        'Programming Language :: Python :: 3 :: Only',
-        'Topic :: Software Development',
-        'Topic :: Scientific/Engineering',
-        "Topic :: Scientific/Engineering :: Artificial Intelligence",
-        "Topic :: Scientific/Engineering :: Image Recognition",
-        'Typing :: Typed',
-        'Operating System :: Microsoft :: Windows',
-        'Operating System :: POSIX',
-        'Operating System :: Unix',
-        'Operating System :: MacOS'
-    ],
-    keywords="machine-learning, deep-learning, vision, ML, DL, AI, YOLOv5, YOLOv8, Roboflow",
-    python_requires='>=3.7',
-)

supervision/__init__.py

@@ -1,17 +1,41 @@
-__version__ = "0.5.0"
+import importlib.metadata as importlib_metadata
 
-from supervision.annotation.voc import detections_to_voc_xml
+try:
+    # This will read version from pyproject.toml
+    __version__ = importlib_metadata.version(__package__ or __name__)
+except importlib_metadata.PackageNotFoundError:
+    __version__ = "development"
+
+
+from supervision.classification.core import Classifications
+from supervision.dataset.core import (
+    BaseDataset,
+    ClassificationDataset,
+    DetectionDataset,
+)
 from supervision.detection.annotate import BoxAnnotator, MaskAnnotator
 from supervision.detection.core import Detections
 from supervision.detection.line_counter import LineZone, LineZoneAnnotator
 from supervision.detection.tools.polygon_zone import PolygonZone, PolygonZoneAnnotator
-from supervision.detection.utils import generate_2d_mask, mask_to_xyxy
+from supervision.detection.utils import (
+    box_iou_batch,
+    filter_polygons_by_area,
+    mask_to_polygons,
+    mask_to_xyxy,
+    non_max_suppression,
+    polygon_to_mask,
+    polygon_to_xyxy,
+)
 from supervision.draw.color import Color, ColorPalette
 from supervision.draw.utils import draw_filled_rectangle, draw_polygon, draw_text
 from supervision.geometry.core import Point, Position, Rect
 from supervision.geometry.utils import get_polygon_center
-from supervision.notebook.utils import plot_image, plot_images_grid
-from supervision.video import (
+from supervision.metrics.detection import ConfusionMatrix, MeanAveragePrecision
+from supervision.tracker.byte_tracker.core import ByteTrack
+from supervision.utils.file import list_files_with_extensions
+from supervision.utils.image import ImageSink, crop
+from supervision.utils.notebook import plot_image, plot_images_grid
+from supervision.utils.video import (
     VideoInfo,
     VideoSink,
     get_video_frames_generator,

supervision/annotation/voc.py

@@ -1,104 +0,0 @@
-from typing import List
-from xml.dom.minidom import parseString
-from xml.etree.ElementTree import Element, SubElement, tostring
-
-from supervision.detection.core import Detections
-
-
-def detections_to_voc_xml(
-    detections: Detections,
-    classes: List[str],
-    filename: str,
-    width: int,
-    height: int,
-    depth: int = 3,
-) -> str:
-    """
-    Converts Detections object to Pascal VOC XML format.
-
-    Args:
-        detections (Detections): A Detections object containing bounding boxes, class ids, and other relevant information.
-        classes (List[str]): A list of class names corresponding to the class ids in the Detections object.
-        filename (str): The name of the image file associated with the detections.
-        width (int): The width of the image in pixels.
-        height (int): The height of the image in pixels.
-        depth (int, optional): The number of color channels in the image. Defaults to 3 for RGB images.
-
-    Returns:
-        str: An XML string in Pascal VOC format representing the detections.
-
-    Examples:
-        ```python
-        >>> import numpy as np
-        >>> import supervision as sv
-
-        >>> xyxy = np.array([
-        ...     [50, 30, 200, 180],
-        ...     [20, 40, 150, 190]
-        ... ])
-        >>> class_id = np.array([1, 0])
-        >>> detections = Detections(xyxy=xyxy, class_id=class_id)
-
-        >>> classes = ["dog", "cat"]
-
-        >>> voc_xml = detections_to_voc_xml(
-        ...     detections=detections,
-        ...     classes=classes,
-        ...     filename="image1.jpg",
-        ...     width=500,
-        ...     height=400
-        ... )
-        ```
-    """
-
-    # Create root element
-    annotation = Element("annotation")
-
-    # Add folder element
-    folder = SubElement(annotation, "folder")
-    folder.text = "VOC"
-
-    # Add filename element
-    fname = SubElement(annotation, "filename")
-    fname.text = filename
-
-    # Add source element
-    source = SubElement(annotation, "source")
-    database = SubElement(source, "database")
-    database.text = "roboflow.ai"
-
-    # Add size element
-    size = SubElement(annotation, "size")
-    w = SubElement(size, "width")
-    w.text = str(width)
-    h = SubElement(size, "height")
-    h.text = str(height)
-    d = SubElement(size, "depth")
-    d.text = str(depth)
-
-    # Add segmented element
-    segmented = SubElement(annotation, "segmented")
-    segmented.text = "0"
-
-    # Add object elements
-    for i in range(detections.xyxy.shape[0]):
-        obj = SubElement(annotation, "object")
-
-        class_id = detections.class_id[i] if detections.class_id is not None else None
-        name = SubElement(obj, "name")
-        name.text = classes[class_id] if class_id is not None else "unknown"
-
-        bndbox = SubElement(obj, "bndbox")
-        xmin = SubElement(bndbox, "xmin")
-        xmin.text = str(int(detections.xyxy[i, 0]))
-        ymin = SubElement(bndbox, "ymin")
-        ymin.text = str(int(detections.xyxy[i, 1]))
-        xmax = SubElement(bndbox, "xmax")
-        xmax.text = str(int(detections.xyxy[i, 2]))
-        ymax = SubElement(bndbox, "ymax")
-        ymax.text = str(int(detections.xyxy[i, 3]))
-
-    # Generate XML string
-    xml_string = parseString(tostring(annotation)).toprettyxml(indent="  ")
-
-    return xml_string

supervision/classification/core.py

@@ -0,0 +1,97 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Any, Optional, Tuple
+
+import numpy as np
+
+
+def _validate_class_ids(class_id: Any, n: int) -> None:
+    """
+    Ensure that class_id is a 1d np.ndarray with (n, ) shape.
+    """
+    is_valid = isinstance(class_id, np.ndarray) and class_id.shape == (n,)
+    if not is_valid:
+        raise ValueError("class_id must be 1d np.ndarray with (n, ) shape")
+
+
+def _validate_confidence(confidence: Any, n: int) -> None:
+    """
+    Ensure that confidence is a 1d np.ndarray with (n, ) shape.
+    """
+    if confidence is not None:
+        is_valid = isinstance(confidence, np.ndarray) and confidence.shape == (n,)
+        if not is_valid:
+            raise ValueError("confidence must be 1d np.ndarray with (n, ) shape")
+
+
+@dataclass
+class Classifications:
+    class_id: np.ndarray
+    confidence: Optional[np.ndarray] = None
+
+    def __post_init__(self) -> None:
+        """
+        Validate the classification inputs.
+        """
+        n = len(self.class_id)
+
+        _validate_class_ids(self.class_id, n)
+        _validate_confidence(self.confidence, n)
+
+    @classmethod
+    def from_yolov8(cls, yolov8_results) -> Classifications:
+        """
+        Creates a Classifications instance from a [YOLOv8](https://github.com/ultralytics/ultralytics) inference result.
+
+        Args:
+            yolov8_results (ultralytics.yolo.engine.results.Results): The output Results instance from YOLOv8
+
+        Returns:
+            Detections: A new Classifications object.
+
+        Example:
+            ```python
+            >>> import cv2
+            >>> from ultralytics import YOLO
+            >>> import supervision as sv
+
+            >>> image = cv2.imread(SOURCE_IMAGE_PATH)
+            >>> model = YOLO('yolov8s-cls.pt')
+            >>> result = model(image)[0]
+            >>> classifications = sv.Classifications.from_yolov8(result)
+            ```
+        """
+        confidence = yolov8_results.probs.data.cpu().numpy()
+        return cls(class_id=np.arange(confidence.shape[0]), confidence=confidence)
+
+    def get_top_k(self, k: int) -> Tuple[np.ndarray, np.ndarray]:
+        """
+        Retrieve the top k class IDs and confidences, ordered in descending order by confidence.
+
+        Args:
+            k (int): The number of top class IDs and confidences to retrieve.
+
+        Returns:
+            Tuple[np.ndarray, np.ndarray]: A tuple containing the top k class IDs and confidences.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> classifications = sv.Classifications(...)
+
+            >>> classifications.get_top_k(1)
+
+            (array([1]), array([0.9]))
+            ```
+        """
+        if self.confidence is None:
+            raise ValueError("top_k could not be calculated, confidence is None")
+
+        order = np.argsort(self.confidence)[::-1]
+        top_k_order = order[:k]
+        top_k_class_id = self.class_id[top_k_order]
+        top_k_confidence = self.confidence[top_k_order]
+
+        return top_k_class_id, top_k_confidence

supervision/dataset/core.py

@@ -0,0 +1,624 @@
+from __future__ import annotations
+
+import os
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Dict, Iterator, List, Optional, Tuple
+
+import cv2
+import numpy as np
+
+from supervision.classification.core import Classifications
+from supervision.dataset.formats.coco import (
+    load_coco_annotations,
+    save_coco_annotations,
+)
+from supervision.dataset.formats.pascal_voc import (
+    detections_to_pascal_voc,
+    load_pascal_voc_annotations,
+)
+from supervision.dataset.formats.yolo import (
+    load_yolo_annotations,
+    save_data_yaml,
+    save_yolo_annotations,
+)
+from supervision.dataset.utils import (
+    build_class_index_mapping,
+    map_detections_class_id,
+    merge_class_lists,
+    save_dataset_images,
+    train_test_split,
+)
+from supervision.detection.core import Detections
+
+
+@dataclass
+class BaseDataset(ABC):
+    @abstractmethod
+    def __len__(self) -> int:
+        pass
+
+    @abstractmethod
+    def split(
+        self, split_ratio=0.8, random_state=None, shuffle: bool = True
+    ) -> Tuple[BaseDataset, BaseDataset]:
+        pass
+
+
+@dataclass
+class DetectionDataset(BaseDataset):
+    """
+    Dataclass containing information about object detection dataset.
+
+    Attributes:
+        classes (List[str]): List containing dataset class names.
+        images (Dict[str, np.ndarray]): Dictionary mapping image name to image.
+        annotations (Dict[str, Detections]): Dictionary mapping image name to annotations.
+    """
+
+    classes: List[str]
+    images: Dict[str, np.ndarray]
+    annotations: Dict[str, Detections]
+
+    def __len__(self) -> int:
+        """
+        Return the number of images in the dataset.
+
+        Returns:
+            int: The number of images.
+        """
+        return len(self.images)
+
+    def __iter__(self) -> Iterator[Tuple[str, np.ndarray, Detections]]:
+        """
+        Iterate over the images and annotations in the dataset.
+
+        Yields:
+            Iterator[Tuple[str, np.ndarray, Detections]]: An iterator that yields tuples containing the image name,
+                                                          the image data, and its corresponding annotation.
+        """
+        for image_name, image in self.images.items():
+            yield image_name, image, self.annotations.get(image_name, None)
+
+    def __eq__(self, other):
+        if not isinstance(other, DetectionDataset):
+            return False
+
+        if set(self.classes) != set(other.classes):
+            return False
+
+        for key in self.images:
+            if not np.array_equal(self.images[key], other.images[key]):
+                return False
+            if not self.annotations[key] == other.annotations[key]:
+                return False
+
+        return True
+
+    def split(
+        self, split_ratio=0.8, random_state=None, shuffle: bool = True
+    ) -> Tuple[DetectionDataset, DetectionDataset]:
+        """
+        Splits the dataset into two parts (training and testing) using the provided split_ratio.
+
+        Args:
+            split_ratio (float, optional): The ratio of the training set to the entire dataset.
+            random_state (int, optional): The seed for the random number generator. This is used for reproducibility.
+            shuffle (bool, optional): Whether to shuffle the data before splitting.
+
+        Returns:
+            Tuple[DetectionDataset, DetectionDataset]: A tuple containing the training and testing datasets.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> ds = sv.DetectionDataset(...)
+            >>> train_ds, test_ds = ds.split(split_ratio=0.7, random_state=42, shuffle=True)
+            >>> len(train_ds), len(test_ds)
+            (700, 300)
+            ```
+        """
+
+        image_names = list(self.images.keys())
+        train_names, test_names = train_test_split(
+            data=image_names,
+            train_ratio=split_ratio,
+            random_state=random_state,
+            shuffle=shuffle,
+        )
+
+        train_dataset = DetectionDataset(
+            classes=self.classes,
+            images={name: self.images[name] for name in train_names},
+            annotations={name: self.annotations[name] for name in train_names},
+        )
+        test_dataset = DetectionDataset(
+            classes=self.classes,
+            images={name: self.images[name] for name in test_names},
+            annotations={name: self.annotations[name] for name in test_names},
+        )
+        return train_dataset, test_dataset
+
+    def as_pascal_voc(
+        self,
+        images_directory_path: Optional[str] = None,
+        annotations_directory_path: Optional[str] = None,
+        min_image_area_percentage: float = 0.0,
+        max_image_area_percentage: float = 1.0,
+        approximation_percentage: float = 0.0,
+    ) -> None:
+        """
+        Exports the dataset to PASCAL VOC format. This method saves the images and their corresponding annotations in
+        PASCAL VOC format.
+
+        Args:
+            images_directory_path (Optional[str]): The path to the directory where the images should be saved.
+                If not provided, images will not be saved.
+            annotations_directory_path (Optional[str]): The path to the directory where the annotations in
+                PASCAL VOC format should be saved. If not provided, annotations will not be saved.
+            min_image_area_percentage (float): The minimum percentage of detection area relative to
+                the image area for a detection to be included. Argument is used only for segmentation datasets.
+            max_image_area_percentage (float): The maximum percentage of detection area relative to
+                the image area for a detection to be included. Argument is used only for segmentation datasets.
+            approximation_percentage (float): The percentage of polygon points to be removed from the input polygon,
+            in the range [0, 1). Argument is used only for segmentation datasets.
+        """
+        if images_directory_path:
+            images_path = Path(images_directory_path)
+            images_path.mkdir(parents=True, exist_ok=True)
+
+        if annotations_directory_path:
+            annotations_path = Path(annotations_directory_path)
+            annotations_path.mkdir(parents=True, exist_ok=True)
+
+        for image_name, image in self.images.items():
+            detections = self.annotations[image_name]
+
+            if images_directory_path:
+                cv2.imwrite(str(images_path / image_name), image)
+
+            if annotations_directory_path:
+                annotation_name = Path(image_name).stem
+                pascal_voc_xml = detections_to_pascal_voc(
+                    detections=detections,
+                    classes=self.classes,
+                    filename=image_name,
+                    image_shape=image.shape,
+                    min_image_area_percentage=min_image_area_percentage,
+                    max_image_area_percentage=max_image_area_percentage,
+                    approximation_percentage=approximation_percentage,
+                )
+
+                with open(annotations_path / f"{annotation_name}.xml", "w") as f:
+                    f.write(pascal_voc_xml)
+
+    @classmethod
+    def from_pascal_voc(
+        cls,
+        images_directory_path: str,
+        annotations_directory_path: str,
+        force_masks: bool = False,
+    ) -> DetectionDataset:
+        """
+        Creates a Dataset instance from PASCAL VOC formatted data.
+
+        Args:
+            images_directory_path (str): The path to the directory containing the images.
+            annotations_directory_path (str): The path to the directory containing the PASCAL VOC XML annotations.
+            force_masks (bool, optional): If True, forces masks to be loaded for all annotations, regardless of whether they are present.
+
+        Returns:
+            DetectionDataset: A DetectionDataset instance containing the loaded images and annotations.
+
+        Example:
+            ```python
+            >>> import roboflow
+            >>> from roboflow import Roboflow
+            >>> import supervision as sv
+
+            >>> roboflow.login()
+
+            >>> rf = Roboflow()
+
+            >>> project = rf.workspace(WORKSPACE_ID).project(PROJECT_ID)
+            >>> dataset = project.version(PROJECT_VERSION).download("voc")
+
+            >>> ds = sv.DetectionDataset.from_pascal_voc(
+            ...     images_directory_path=f"{dataset.location}/train/images",
+            ...     annotations_directory_path=f"{dataset.location}/train/labels"
+            ... )
+
+            >>> ds.classes
+            ['dog', 'person']
+            ```
+        """
+
+        classes, images, annotations = load_pascal_voc_annotations(
+            images_directory_path=images_directory_path,
+            annotations_directory_path=annotations_directory_path,
+            force_masks=force_masks,
+        )
+
+        return DetectionDataset(classes=classes, images=images, annotations=annotations)
+
+    @classmethod
+    def from_yolo(
+        cls,
+        images_directory_path: str,
+        annotations_directory_path: str,
+        data_yaml_path: str,
+        force_masks: bool = False,
+    ) -> DetectionDataset:
+        """
+        Creates a Dataset instance from YOLO formatted data.
+
+        Args:
+            images_directory_path (str): The path to the directory containing the images.
+            annotations_directory_path (str): The path to the directory containing the YOLO annotation files.
+            data_yaml_path (str): The path to the data 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.
+
+        Returns:
+            DetectionDataset: A DetectionDataset instance containing the loaded images and annotations.
+
+        Example:
+            ```python
+            >>> import roboflow
+            >>> from roboflow import Roboflow
+            >>> import supervision as sv
+
+            >>> roboflow.login()
+
+            >>> rf = Roboflow()
+
+            >>> project = rf.workspace(WORKSPACE_ID).project(PROJECT_ID)
+            >>> dataset = project.version(PROJECT_VERSION).download("yolov5")
+
+            >>> ds = sv.DetectionDataset.from_yolo(
+            ...     images_directory_path=f"{dataset.location}/train/images",
+            ...     annotations_directory_path=f"{dataset.location}/train/labels",
+            ...     data_yaml_path=f"{dataset.location}/data.yaml"
+            ... )
+
+            >>> ds.classes
+            ['dog', 'person']
+            ```
+        """
+        classes, images, annotations = load_yolo_annotations(
+            images_directory_path=images_directory_path,
+            annotations_directory_path=annotations_directory_path,
+            data_yaml_path=data_yaml_path,
+            force_masks=force_masks,
+        )
+        return DetectionDataset(classes=classes, images=images, annotations=annotations)
+
+    def as_yolo(
+        self,
+        images_directory_path: Optional[str] = None,
+        annotations_directory_path: Optional[str] = None,
+        data_yaml_path: Optional[str] = None,
+        min_image_area_percentage: float = 0.0,
+        max_image_area_percentage: float = 1.0,
+        approximation_percentage: float = 0.0,
+    ) -> None:
+        """
+        Exports the dataset to YOLO format. This method saves the images and their corresponding
+        annotations in YOLO format.
+
+        Args:
+            images_directory_path (Optional[str]): The path to the directory where the images should be saved.
+                If not provided, images will not be saved.
+            annotations_directory_path (Optional[str]): The path to the directory where the annotations in
+                YOLO format should be saved. If not provided, annotations will not be saved.
+            data_yaml_path (Optional[str]): The path where the data.yaml file should be saved.
+                If not provided, the file will not be saved.
+            min_image_area_percentage (float): The minimum percentage of detection area relative to
+                the image area for a detection to be included. Argument is used only for segmentation datasets.
+            max_image_area_percentage (float): The maximum percentage of detection area relative to
+                the image area for a detection to be included. Argument is used only for segmentation datasets.
+            approximation_percentage (float): The percentage of polygon points to be removed from the input polygon,
+                in the range [0, 1). This is useful for simplifying the annotations. Argument is used only for
+                segmentation datasets.
+        """
+        if images_directory_path is not None:
+            save_dataset_images(
+                images_directory_path=images_directory_path, images=self.images
+            )
+        if annotations_directory_path is not None:
+            save_yolo_annotations(
+                annotations_directory_path=annotations_directory_path,
+                images=self.images,
+                annotations=self.annotations,
+                min_image_area_percentage=min_image_area_percentage,
+                max_image_area_percentage=max_image_area_percentage,
+                approximation_percentage=approximation_percentage,
+            )
+        if data_yaml_path is not None:
+            save_data_yaml(data_yaml_path=data_yaml_path, classes=self.classes)
+
+    @classmethod
+    def from_coco(
+        cls,
+        images_directory_path: str,
+        annotations_path: str,
+        force_masks: bool = False,
+    ) -> DetectionDataset:
+        """
+        Creates a Dataset instance from COCO formatted data.
+
+        Args:
+            images_directory_path (str): The path to the directory containing the images.
+            annotations_path (str): The path to the json annotation files.
+            force_masks (bool, optional): If True, forces masks to be loaded for all annotations, regardless of whether they are present.
+
+        Returns:
+            DetectionDataset: A DetectionDataset instance containing the loaded images and annotations.
+
+        Example:
+            ```python
+            >>> import roboflow
+            >>> from roboflow import Roboflow
+            >>> import supervision as sv
+
+            >>> roboflow.login()
+
+            >>> rf = Roboflow()
+
+            >>> project = rf.workspace(WORKSPACE_ID).project(PROJECT_ID)
+            >>> dataset = project.version(PROJECT_VERSION).download("coco")
+
+            >>> ds = sv.DetectionDataset.from_coco(
+            ...     images_directory_path=f"{dataset.location}/train",
+            ...     annotations_path=f"{dataset.location}/train/_annotations.coco.json",
+            ... )
+
+            >>> ds.classes
+            ['dog', 'person']
+            ```
+        """
+        classes, images, annotations = load_coco_annotations(
+            images_directory_path=images_directory_path,
+            annotations_path=annotations_path,
+            force_masks=force_masks,
+        )
+        return DetectionDataset(classes=classes, images=images, annotations=annotations)
+
+    def as_coco(
+        self,
+        images_directory_path: Optional[str] = None,
+        annotations_path: Optional[str] = None,
+        min_image_area_percentage: float = 0.0,
+        max_image_area_percentage: float = 1.0,
+        approximation_percentage: float = 0.0,
+    ) -> None:
+        """
+        Exports the dataset to COCO format. This method saves the images and their corresponding
+        annotations in COCO format.
+
+        Args:
+            images_directory_path (Optional[str]): The path to the directory where the images should be saved.
+                If not provided, images will not be saved.
+            annotations_path (Optional[str]): The path to COCO annotation file.
+            min_image_area_percentage (float): The minimum percentage of detection area relative to
+                the image area for a detection to be included. Argument is used only for segmentation datasets.
+            max_image_area_percentage (float): The maximum percentage of detection area relative to
+                the image area for a detection to be included. Argument is used only for segmentation datasets.
+            approximation_percentage (float): The percentage of polygon points to be removed from the input polygon,
+                in the range [0, 1). This is useful for simplifying the annotations. Argument is used only for
+                segmentation datasets.
+        """
+        if images_directory_path is not None:
+            save_dataset_images(
+                images_directory_path=images_directory_path, images=self.images
+            )
+        if annotations_path is not None:
+            save_coco_annotations(
+                annotation_path=annotations_path,
+                images=self.images,
+                annotations=self.annotations,
+                classes=self.classes,
+                min_image_area_percentage=min_image_area_percentage,
+                max_image_area_percentage=max_image_area_percentage,
+                approximation_percentage=approximation_percentage,
+            )
+
+    @classmethod
+    def merge(cls, dataset_list: List[DetectionDataset]) -> DetectionDataset:
+        """
+        Merge a list of `DetectionDataset` objects into a single `DetectionDataset` object.
+
+        This method takes a list of `DetectionDataset` objects and combines their respective fields (`classes`, `images`,
+        `annotations`) into a single `DetectionDataset` object.
+
+        Args:
+            dataset_list (List[DetectionDataset]): A list of `DetectionDataset` objects to merge.
+
+        Returns:
+            (DetectionDataset): A single `DetectionDataset` object containing the merged data from the input list.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> ds_1 = sv.DetectionDataset(...)
+            >>> len(ds_1)
+            100
+            >>> ds_1.classes
+            ['dog', 'person']
+
+            >>> ds_2 = sv.DetectionDataset(...)
+            >>> len(ds_2)
+            200
+            >>> ds_2.classes
+            ['cat']
+
+            >>> ds_merged = sv.DetectionDataset.merge([ds_1, ds_2])
+            >>> len(ds_merged)
+            300
+            >>> ds_merged.classes
+            ['cat', 'dog', 'person']
+            ```
+        """
+        merged_images, merged_annotations = {}, {}
+        class_lists = [dataset.classes for dataset in dataset_list]
+        merged_classes = merge_class_lists(class_lists=class_lists)
+
+        for dataset in dataset_list:
+            class_index_mapping = build_class_index_mapping(
+                source_classes=dataset.classes, target_classes=merged_classes
+            )
+            for image_name, image, detections in dataset:
+                if image_name in merged_annotations:
+                    raise ValueError(
+                        f"Image name {image_name} is not unique across datasets."
+                    )
+
+                merged_images[image_name] = image
+                merged_annotations[image_name] = map_detections_class_id(
+                    source_to_target_mapping=class_index_mapping,
+                    detections=detections,
+                )
+
+        return cls(
+            classes=merged_classes, images=merged_images, annotations=merged_annotations
+        )
+
+
+@dataclass
+class ClassificationDataset(BaseDataset):
+    """
+    Dataclass containing information about a classification dataset.
+
+    Attributes:
+        classes (List[str]): List containing dataset class names.
+        images (Dict[str, np.ndarray]): Dictionary mapping image name to image.
+        annotations (Dict[str, Detections]): Dictionary mapping image name to annotations.
+    """
+
+    classes: List[str]
+    images: Dict[str, np.ndarray]
+    annotations: Dict[str, Classifications]
+
+    def __len__(self) -> int:
+        return len(self.images)
+
+    def split(
+        self, split_ratio=0.8, random_state=None, shuffle: bool = True
+    ) -> Tuple[ClassificationDataset, ClassificationDataset]:
+        """
+        Splits the dataset into two parts (training and testing) using the provided split_ratio.
+
+        Args:
+            split_ratio (float, optional): The ratio of the training set to the entire dataset.
+            random_state (int, optional): The seed for the random number generator. This is used for reproducibility.
+            shuffle (bool, optional): Whether to shuffle the data before splitting.
+
+        Returns:
+            Tuple[ClassificationDataset, ClassificationDataset]: A tuple containing the training and testing datasets.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> cd = sv.ClassificationDataset(...)
+            >>> train_cd, test_cd = cd.split(split_ratio=0.7, random_state=42, shuffle=True)
+            >>> len(train_cd), len(test_cd)
+            (700, 300)
+            ```
+        """
+        image_names = list(self.images.keys())
+        train_names, test_names = train_test_split(
+            data=image_names,
+            train_ratio=split_ratio,
+            random_state=random_state,
+            shuffle=shuffle,
+        )
+
+        train_dataset = ClassificationDataset(
+            classes=self.classes,
+            images={name: self.images[name] for name in train_names},
+            annotations={name: self.annotations[name] for name in train_names},
+        )
+        test_dataset = ClassificationDataset(
+            classes=self.classes,
+            images={name: self.images[name] for name in test_names},
+            annotations={name: self.annotations[name] for name in test_names},
+        )
+        return train_dataset, test_dataset
+
+    def as_folder_structure(self, root_directory_path: str) -> None:
+        """
+        Saves the dataset as a multi-class folder structure.
+
+        Args:
+            root_directory_path (str): The path to the directory where the dataset will be saved.
+        """
+        os.makedirs(root_directory_path, exist_ok=True)
+
+        for class_name in self.classes:
+            os.makedirs(os.path.join(root_directory_path, class_name), exist_ok=True)
+
+        for image_name in self.images:
+            classification = self.annotations[image_name]
+            image = self.images[image_name]
+            class_id = (
+                classification.class_id[0]
+                if classification.confidence is None
+                else classification.get_top_k(1)[0][0]
+            )
+            class_name = self.classes[class_id]
+            image_path = os.path.join(root_directory_path, class_name, image_name)
+            cv2.imwrite(image_path, image)
+
+    @classmethod
+    def from_folder_structure(cls, root_directory_path: str) -> ClassificationDataset:
+        """
+        Load data from a multiclass folder structure into a ClassificationDataset.
+
+        Args:
+            root_directory_path (str): The path to the dataset directory.
+
+        Returns:
+            ClassificationDataset: The dataset.
+
+        Example:
+            ```python
+            >>> import roboflow
+            >>> from roboflow import Roboflow
+            >>> import supervision as sv
+
+            >>> roboflow.login()
+
+            >>> rf = Roboflow()
+
+            >>> project = rf.workspace(WORKSPACE_ID).project(PROJECT_ID)
+            >>> dataset = project.version(PROJECT_VERSION).download("folder")
+
+            >>> cd = sv.ClassificationDataset.from_folder_structure(
+            ...     root_directory_path=f"{dataset.location}/train"
+            ... )
+            ```
+        """
+        classes = os.listdir(root_directory_path)
+        classes = sorted(set(classes))
+
+        images = {}
+        annotations = {}
+
+        for class_name in classes:
+            class_id = classes.index(class_name)
+
+            for image in os.listdir(os.path.join(root_directory_path, class_name)):
+                image_dir = os.path.join(root_directory_path, class_name, image)
+                images[image] = cv2.imread(image_dir)
+                annotations[image] = Classifications(
+                    class_id=np.array([class_id]),
+                )
+
+        return cls(
+            classes=classes,
+            images=images,
+            annotations=annotations,
+        )

supervision/dataset/formats/coco.py

@@ -0,0 +1,237 @@
+import os
+from datetime import datetime
+from pathlib import Path
+from typing import Dict, List, Tuple
+
+import cv2
+import numpy as np
+
+from supervision.dataset.utils import (
+    approximate_mask_with_polygons,
+    map_detections_class_id,
+)
+from supervision.detection.core import Detections
+from supervision.detection.utils import polygon_to_mask
+from supervision.utils.file import read_json_file, save_json_file
+
+
+def coco_categories_to_classes(coco_categories: List[dict]) -> List[str]:
+    return [
+        category["name"]
+        for category in sorted(coco_categories, key=lambda category: category["id"])
+    ]
+
+
+def build_coco_class_index_mapping(
+    coco_categories: List[dict], target_classes: List[str]
+) -> Dict[int, int]:
+    source_class_to_index = {
+        category["name"]: category["id"] for category in coco_categories
+    }
+    return {
+        source_class_to_index[target_class_name]: target_class_index
+        for target_class_index, target_class_name in enumerate(target_classes)
+    }
+
+
+def classes_to_coco_categories(classes: List[str]) -> List[dict]:
+    return [
+        {
+            "id": class_id,
+            "name": class_name,
+            "supercategory": "common-objects",
+        }
+        for class_id, class_name in enumerate(classes)
+    ]
+
+
+def group_coco_annotations_by_image_id(
+    coco_annotations: List[dict],
+) -> Dict[int, List[dict]]:
+    annotations = {}
+    for annotation in coco_annotations:
+        image_id = annotation["image_id"]
+        if image_id not in annotations:
+            annotations[image_id] = []
+        annotations[image_id].append(annotation)
+    return annotations
+
+
+def _polygons_to_masks(
+    polygons: List[np.ndarray], resolution_wh: Tuple[int, int]
+) -> np.ndarray:
+    return np.array(
+        [
+            polygon_to_mask(polygon=polygon, resolution_wh=resolution_wh)
+            for polygon in polygons
+        ],
+        dtype=bool,
+    )
+
+
+def coco_annotations_to_detections(
+    image_annotations: List[dict], resolution_wh: Tuple[int, int], with_masks: bool
+) -> Detections:
+    if not image_annotations:
+        return Detections.empty()
+
+    class_ids = [
+        image_annotation["category_id"] for image_annotation in image_annotations
+    ]
+    xyxy = [image_annotation["bbox"] for image_annotation in image_annotations]
+    xyxy = np.asarray(xyxy)
+    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)
+        return Detections(
+            class_id=np.asarray(class_ids, dtype=int), xyxy=xyxy, mask=mask
+        )
+
+    return Detections(xyxy=xyxy, class_id=np.asarray(class_ids, dtype=int))
+
+
+def detections_to_coco_annotations(
+    detections: Detections,
+    image_id: int,
+    annotation_id: int,
+    min_image_area_percentage: float = 0.0,
+    max_image_area_percentage: float = 1.0,
+    approximation_percentage: float = 0.75,
+) -> Tuple[List[Dict], int]:
+    coco_annotations = []
+    for xyxy, mask, _, class_id, _ in detections:
+        box_width, box_height = xyxy[2] - xyxy[0], xyxy[3] - xyxy[1]
+        polygon = []
+        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()
+            )
+        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,
+        }
+        coco_annotations.append(coco_annotation)
+        annotation_id += 1
+    return coco_annotations, annotation_id
+
+
+def load_coco_annotations(
+    images_directory_path: str,
+    annotations_path: str,
+    force_masks: bool = False,
+) -> Tuple[List[str], Dict[str, np.ndarray], Dict[str, Detections]]:
+    coco_data = read_json_file(file_path=annotations_path)
+    classes = coco_categories_to_classes(coco_categories=coco_data["categories"])
+    class_index_mapping = build_coco_class_index_mapping(
+        coco_categories=coco_data["categories"], target_classes=classes
+    )
+    coco_images = coco_data["images"]
+    coco_annotations_groups = group_coco_annotations_by_image_id(
+        coco_annotations=coco_data["annotations"]
+    )
+
+    images = {}
+    annotations = {}
+
+    for coco_image in coco_images:
+        image_name, image_width, image_height = (
+            coco_image["file_name"],
+            coco_image["width"],
+            coco_image["height"],
+        )
+        image_annotations = coco_annotations_groups.get(coco_image["id"], [])
+        image_path = os.path.join(images_directory_path, image_name)
+
+        image = cv2.imread(str(image_path))
+        annotation = coco_annotations_to_detections(
+            image_annotations=image_annotations,
+            resolution_wh=(image_width, image_height),
+            with_masks=force_masks,
+        )
+        annotation = map_detections_class_id(
+            source_to_target_mapping=class_index_mapping,
+            detections=annotation,
+        )
+
+        images[image_name] = image
+        annotations[image_name] = annotation
+
+    return classes, images, annotations
+
+
+def save_coco_annotations(
+    annotation_path: str,
+    images: Dict[str, np.ndarray],
+    annotations: Dict[str, Detections],
+    classes: List[str],
+    min_image_area_percentage: float = 0.0,
+    max_image_area_percentage: float = 1.0,
+    approximation_percentage: float = 0.75,
+) -> None:
+    Path(annotation_path).parent.mkdir(parents=True, exist_ok=True)
+    info = {}
+    licenses = [
+        {
+            "id": 1,
+            "url": "https://creativecommons.org/licenses/by/4.0/",
+            "name": "CC BY 4.0",
+        }
+    ]
+
+    coco_annotations = []
+    coco_images = []
+    coco_categories = classes_to_coco_categories(classes=classes)
+
+    image_id, annotation_id = 1, 1
+    for image_name, image in images.items():
+        image_height, image_width, _ = image.shape
+
+        coco_image = {
+            "id": image_id,
+            "license": 1,
+            "file_name": image_name,
+            "height": image_height,
+            "width": image_width,
+            "date_captured": datetime.now().strftime("%m/%d/%Y,%H:%M:%S"),
+        }
+
+        coco_images.append(coco_image)
+        detections = annotations[image_name]
+
+        coco_annotation, label_id = detections_to_coco_annotations(
+            detections=detections,
+            image_id=image_id,
+            annotation_id=annotation_id,
+            min_image_area_percentage=min_image_area_percentage,
+            max_image_area_percentage=max_image_area_percentage,
+            approximation_percentage=approximation_percentage,
+        )
+
+        coco_annotations.extend(coco_annotation)
+        image_id += 1
+
+    annotation_dict = {
+        "info": info,
+        "licenses": licenses,
+        "categories": coco_categories,
+        "images": coco_images,
+        "annotations": coco_annotations,
+    }
+    save_json_file(annotation_dict, file_path=annotation_path)

supervision/dataset/formats/pascal_voc.py

@@ -0,0 +1,260 @@
+import os
+from pathlib import Path
+from typing import Dict, List, Optional, Tuple
+from xml.dom.minidom import parseString
+from xml.etree.ElementTree import Element, SubElement, parse, tostring
+
+import cv2
+import numpy as np
+
+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
+from supervision.utils.file import list_files_with_extensions
+
+
+def object_to_pascal_voc(
+    xyxy: np.ndarray, name: str, polygon: Optional[np.ndarray] = None
+) -> Element:
+    root = Element("object")
+
+    object_name = SubElement(root, "name")
+    object_name.text = name
+
+    bndbox = SubElement(root, "bndbox")
+    xmin = SubElement(bndbox, "xmin")
+    xmin.text = str(int(xyxy[0]))
+    ymin = SubElement(bndbox, "ymin")
+    ymin.text = str(int(xyxy[1]))
+    xmax = SubElement(bndbox, "xmax")
+    xmax.text = str(int(xyxy[2]))
+    ymax = SubElement(bndbox, "ymax")
+    ymax.text = str(int(xyxy[3]))
+
+    if polygon is not None:
+        object_polygon = SubElement(root, "polygon")
+        for index, point in enumerate(polygon, start=1):
+            x_coordinate, y_coordinate = point
+            x = SubElement(object_polygon, f"x{index}")
+            x.text = str(x_coordinate)
+            y = SubElement(object_polygon, f"y{index}")
+            y.text = str(y_coordinate)
+
+    return root
+
+
+def detections_to_pascal_voc(
+    detections: Detections,
+    classes: List[str],
+    filename: str,
+    image_shape: Tuple[int, int, int],
+    min_image_area_percentage: float = 0.0,
+    max_image_area_percentage: float = 1.0,
+    approximation_percentage: float = 0.75,
+) -> str:
+    """
+    Converts Detections object to Pascal VOC XML format.
+
+    Args:
+        detections (Detections): A Detections object containing bounding boxes, class ids, and other relevant information.
+        classes (List[str]): A list of class names corresponding to the class ids in the Detections object.
+        filename (str): The name of the image file associated with the detections.
+        image_shape (Tuple[int, int, int]): The shape of the image file associated with the detections.
+        min_image_area_percentage (float): Minimum detection area relative to area of image associated with it.
+        max_image_area_percentage (float): Maximum detection area relative to area of image associated with it.
+        approximation_percentage (float): The percentage of polygon points to be removed from the input polygon, in the range [0, 1).
+    Returns:
+        str: An XML string in Pascal VOC format representing the detections.
+    """
+    height, width, depth = image_shape
+
+    # Create root element
+    annotation = Element("annotation")
+
+    # Add folder element
+    folder = SubElement(annotation, "folder")
+    folder.text = "VOC"
+
+    # Add filename element
+    file_name = SubElement(annotation, "filename")
+    file_name.text = filename
+
+    # Add source element
+    source = SubElement(annotation, "source")
+    database = SubElement(source, "database")
+    database.text = "roboflow.ai"
+
+    # Add size element
+    size = SubElement(annotation, "size")
+    w = SubElement(size, "width")
+    w.text = str(width)
+    h = SubElement(size, "height")
+    h.text = str(height)
+    d = SubElement(size, "depth")
+    d.text = str(depth)
+
+    # Add segmented element
+    segmented = SubElement(annotation, "segmented")
+    segmented.text = "0"
+
+    # Add object elements
+    for xyxy, mask, _, class_id, _ in detections:
+        name = classes[class_id]
+        if mask is not None:
+            polygons = 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,
+            )
+            for polygon in polygons:
+                xyxy = polygon_to_xyxy(polygon=polygon)
+                next_object = object_to_pascal_voc(
+                    xyxy=xyxy, name=name, polygon=polygon
+                )
+                annotation.append(next_object)
+        else:
+            next_object = object_to_pascal_voc(xyxy=xyxy, name=name)
+            annotation.append(next_object)
+
+    # Generate XML string
+    xml_string = parseString(tostring(annotation)).toprettyxml(indent="  ")
+
+    return xml_string
+
+
+def load_pascal_voc_annotations(
+    images_directory_path: str,
+    annotations_directory_path: str,
+    force_masks: bool = False,
+) -> Tuple[List[str], Dict[str, np.ndarray], Dict[str, Detections]]:
+    """
+    Loads PASCAL VOC annotations and returns class names, images, and their corresponding detections.
+
+    Args:
+        images_directory_path (str): The path to the directory containing the images.
+        annotations_directory_path (str): The path to the directory containing the PASCAL VOC annotation files.
+        force_masks (bool, optional): If True, forces masks to be loaded for all annotations, regardless of whether they are present.
+
+    Returns:
+        Tuple[List[str], Dict[str, np.ndarray], Dict[str, Detections]]: A tuple containing a list of class names, a dictionary with image names as keys and images as values, and a dictionary with image names as keys and corresponding Detections instances as values.
+    """
+
+    image_paths = list_files_with_extensions(
+        directory=images_directory_path, extensions=["jpg", "jpeg", "png"]
+    )
+
+    classes = []
+    images = {}
+    annotations = {}
+
+    for image_path in image_paths:
+        image_name = Path(image_path).stem
+        image = cv2.imread(str(image_path))
+
+        annotation_path = os.path.join(annotations_directory_path, f"{image_name}.xml")
+        if not os.path.exists(annotation_path):
+            images[image_path.name] = image
+            annotations[image_path.name] = Detections.empty()
+            continue
+
+        tree = parse(annotation_path)
+        root = tree.getroot()
+
+        resolution_wh = (image.shape[1], image.shape[0])
+        annotation, classes = detections_from_xml_obj(
+            root, classes, resolution_wh, force_masks
+        )
+
+        images[image_path.name] = image
+        annotations[image_path.name] = annotation
+
+    return classes, images, annotations
+
+
+def detections_from_xml_obj(
+    root: Element, classes: List[str], resolution_wh, force_masks: bool = False
+) -> Tuple[Detections, List[str]]:
+    """
+    Converts an XML object in Pascal VOC format to a Detections object.
+    Expected XML format:
+    <annotation>
+        ...
+        <object>
+            <name>dog</name>
+            <bndbox>
+                <xmin>48</xmin>
+                <ymin>240</ymin>
+                <xmax>195</xmax>
+                <ymax>371</ymax>
+            </bndbox>
+            <polygon>
+                <x1>48</x1>
+                <y1>240</y1>
+                <x2>195</x2>
+                <y2>240</y2>
+                <x3>195</x3>
+                <y3>371</y3>
+                <x4>48</x4>
+                <y4>371</y4>
+            </polygon>
+        </object>
+    </annotation>
+
+    Returns:
+        Tuple[Detections, List[str]]: A tuple containing a Detections object and an updated list of class names, extended with the class names from the XML object.
+    """
+    xyxy = []
+    class_names = []
+    masks = []
+    with_masks = False
+    extended_classes = classes[:]
+    for obj in root.findall("object"):
+        class_name = obj.find("name").text
+        class_names.append(class_name)
+
+        bbox = obj.find("bndbox")
+        x1 = int(bbox.find("xmin").text)
+        y1 = int(bbox.find("ymin").text)
+        x2 = int(bbox.find("xmax").text)
+        y2 = int(bbox.find("ymax").text)
+
+        xyxy.append([x1, y1, x2, y2])
+
+        with_masks = obj.find("polygon") is not None
+        with_masks = force_masks if force_masks else with_masks
+
+        for polygon in obj.findall("polygon"):
+            polygon_points = parse_polygon_points(polygon)
+
+            mask_from_polygon = polygon_to_mask(
+                polygon=np.array(polygon_points),
+                resolution_wh=resolution_wh,
+            )
+            masks.append(mask_from_polygon)
+
+    xyxy = np.array(xyxy) if len(xyxy) > 0 else np.empty((0, 4))
+    for k in set(class_names):
+        if k not in extended_classes:
+            extended_classes.append(k)
+    class_id = np.array(
+        [extended_classes.index(class_name) for class_name in class_names]
+    )
+
+    if with_masks:
+        annotation = Detections(
+            xyxy=xyxy, mask=np.array(masks).astype(bool), class_id=class_id
+        )
+    else:
+        annotation = Detections(xyxy=xyxy, class_id=class_id)
+    return annotation, extended_classes
+
+
+def parse_polygon_points(polygon: Element) -> List[List[int]]:
+    polygon_points = []
+    coords = polygon.findall(".//*")
+    for i in range(0, len(coords), 2):
+        x = int(coords[i].text)
+        y = int(coords[i + 1].text)
+        polygon_points.append([x, y])
+    return polygon_points

supervision/dataset/formats/yolo.py

@@ -0,0 +1,242 @@
+import os
+from pathlib import Path
+from typing import Dict, List, Optional, Tuple, Union
+
+import cv2
+import numpy as np
+
+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
+from supervision.utils.file import (
+    list_files_with_extensions,
+    read_txt_file,
+    read_yaml_file,
+    save_text_file,
+    save_yaml_file,
+)
+
+
+def _parse_box(values: List[str]) -> np.ndarray:
+    x_center, y_center, width, height = values
+    return np.array(
+        [
+            float(x_center) - float(width) / 2,
+            float(y_center) - float(height) / 2,
+            float(x_center) + float(width) / 2,
+            float(y_center) + float(height) / 2,
+        ],
+        dtype=np.float32,
+    )
+
+
+def _box_to_polygon(box: np.ndarray) -> np.ndarray:
+    return np.array(
+        [[box[0], box[1]], [box[2], box[1]], [box[2], box[3]], [box[0], box[3]]]
+    )
+
+
+def _parse_polygon(values: List[str]) -> np.ndarray:
+    return np.array(values, dtype=np.float32).reshape(-1, 2)
+
+
+def _polygons_to_masks(
+    polygons: List[np.ndarray], resolution_wh: Tuple[int, int]
+) -> np.ndarray:
+    return np.array(
+        [
+            polygon_to_mask(polygon=polygon, resolution_wh=resolution_wh)
+            for polygon in polygons
+        ],
+        dtype=bool,
+    )
+
+
+def _with_mask(lines: List[str]) -> bool:
+    return any([len(line.split()) > 5 for line in lines])
+
+
+def _extract_class_names(file_path: str) -> List[str]:
+    data = read_yaml_file(file_path=file_path)
+    names = data["names"]
+    if isinstance(names, dict):
+        names = [names[key] for key in sorted(names.keys())]
+    return names
+
+
+def _image_name_to_annotation_name(image_name: str) -> str:
+    base_name, _ = os.path.splitext(image_name)
+    return base_name + ".txt"
+
+
+def yolo_annotations_to_detections(
+    lines: List[str], resolution_wh: Tuple[int, int], with_masks: bool
+) -> Detections:
+    if len(lines) == 0:
+        return Detections.empty()
+
+    class_id, relative_xyxy, relative_polygon = [], [], []
+    w, h = resolution_wh
+    for line in lines:
+        values = line.split()
+        class_id.append(int(values[0]))
+        if len(values) == 5:
+            box = _parse_box(values=values[1:])
+            relative_xyxy.append(box)
+            if with_masks:
+                relative_polygon.append(_box_to_polygon(box=box))
+        elif len(values) > 5:
+            polygon = _parse_polygon(values=values[1:])
+            relative_xyxy.append(polygon_to_xyxy(polygon=polygon))
+            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)
+
+    if not with_masks:
+        return Detections(class_id=class_id, xyxy=xyxy)
+
+    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)
+
+
+def load_yolo_annotations(
+    images_directory_path: str,
+    annotations_directory_path: str,
+    data_yaml_path: str,
+    force_masks: bool = False,
+) -> Tuple[List[str], Dict[str, np.ndarray], Dict[str, Detections]]:
+    """
+    Loads YOLO annotations and returns class names, images, and their corresponding detections.
+
+    Args:
+        images_directory_path (str): The path to the directory containing the images.
+        annotations_directory_path (str): The path to the directory containing the YOLO annotation files.
+        data_yaml_path (str): The path to the data 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.
+
+    Returns:
+        Tuple[List[str], Dict[str, np.ndarray], Dict[str, Detections]]: A tuple containing a list of class names, a dictionary with image names as keys and images as values, and a dictionary with image names as keys and corresponding Detections instances as values.
+    """
+    image_paths = list_files_with_extensions(
+        directory=images_directory_path, extensions=["jpg", "jpeg", "png"]
+    )
+
+    classes = _extract_class_names(file_path=data_yaml_path)
+    images = {}
+    annotations = {}
+
+    for image_path in image_paths:
+        image_name = Path(image_path).stem
+        image = cv2.imread(str(image_path))
+
+        annotation_path = os.path.join(annotations_directory_path, f"{image_name}.txt")
+        if not os.path.exists(annotation_path):
+            images[image_path.name] = image
+            annotations[image_path.name] = Detections.empty()
+            continue
+
+        lines = read_txt_file(str(annotation_path))
+        h, w, _ = image.shape
+        resolution_wh = (w, h)
+
+        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
+        )
+
+        images[image_path.name] = image
+        annotations[image_path.name] = annotation
+    return classes, images, annotations
+
+
+def object_to_yolo(
+    xyxy: np.ndarray,
+    class_id: int,
+    image_shape: Tuple[int, int, int],
+    polygon: Optional[np.ndarray] = None,
+) -> str:
+    h, w, _ = image_shape
+    if polygon is None:
+        xyxy_relative = xyxy / np.array([w, h, w, h], dtype=np.float32)
+        x_min, y_min, x_max, y_max = xyxy_relative
+        x_center = (x_min + x_max) / 2
+        y_center = (y_min + y_max) / 2
+        width = x_max - x_min
+        height = y_max - y_min
+        return f"{int(class_id)} {x_center:.5f} {y_center:.5f} {width:.5f} {height:.5f}"
+    else:
+        polygon_relative = polygon / np.array([w, h], dtype=np.float32)
+        polygon_relative = polygon_relative.reshape(-1)
+        polygon_parsed = " ".join([f"{value:.5f}" for value in polygon_relative])
+        return f"{int(class_id)} {polygon_parsed}"
+
+
+def detections_to_yolo_annotations(
+    detections: Detections,
+    image_shape: Tuple[int, int, int],
+    min_image_area_percentage: float = 0.0,
+    max_image_area_percentage: float = 1.0,
+    approximation_percentage: float = 0.75,
+) -> List[str]:
+    annotation = []
+    for xyxy, mask, _, class_id, _ in detections:
+        if mask is not None:
+            polygons = 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,
+            )
+            for polygon in polygons:
+                xyxy = polygon_to_xyxy(polygon=polygon)
+                next_object = object_to_yolo(
+                    xyxy=xyxy,
+                    class_id=class_id,
+                    image_shape=image_shape,
+                    polygon=polygon,
+                )
+                annotation.append(next_object)
+        else:
+            next_object = object_to_yolo(
+                xyxy=xyxy, class_id=class_id, image_shape=image_shape
+            )
+            annotation.append(next_object)
+    return annotation
+
+
+def save_yolo_annotations(
+    annotations_directory_path: str,
+    images: Dict[str, np.ndarray],
+    annotations: Dict[str, Detections],
+    min_image_area_percentage: float = 0.0,
+    max_image_area_percentage: float = 1.0,
+    approximation_percentage: float = 0.75,
+) -> None:
+    Path(annotations_directory_path).mkdir(parents=True, exist_ok=True)
+    for image_name, image in images.items():
+        detections = annotations[image_name]
+        yolo_annotations_name = _image_name_to_annotation_name(image_name=image_name)
+        yolo_annotations_path = os.path.join(
+            annotations_directory_path, yolo_annotations_name
+        )
+        lines = detections_to_yolo_annotations(
+            detections=detections,
+            image_shape=image.shape,
+            min_image_area_percentage=min_image_area_percentage,
+            max_image_area_percentage=max_image_area_percentage,
+            approximation_percentage=approximation_percentage,
+        )
+        save_text_file(lines=lines, file_path=yolo_annotations_path)
+
+
+def save_data_yaml(data_yaml_path: str, classes: List[str]) -> None:
+    data = {"nc": len(classes), "names": classes}
+    Path(data_yaml_path).parent.mkdir(parents=True, exist_ok=True)
+    save_yaml_file(data=data, file_path=data_yaml_path)

supervision/dataset/utils.py

@@ -0,0 +1,130 @@
+import copy
+import os
+import random
+from pathlib import Path
+from typing import Dict, List, Optional, Tuple, TypeVar
+
+import cv2
+import numpy as np
+
+from supervision.detection.core import Detections
+from supervision.detection.utils import (
+    approximate_polygon,
+    filter_polygons_by_area,
+    mask_to_polygons,
+)
+
+T = TypeVar("T")
+
+
+def approximate_mask_with_polygons(
+    mask: np.ndarray,
+    min_image_area_percentage: float = 0.0,
+    max_image_area_percentage: float = 1.0,
+    approximation_percentage: float = 0.75,
+) -> List[np.ndarray]:
+    height, width = mask.shape
+    image_area = height * width
+    minimum_detection_area = min_image_area_percentage * image_area
+    maximum_detection_area = max_image_area_percentage * image_area
+
+    polygons = mask_to_polygons(mask=mask)
+    if len(polygons) == 1:
+        polygons = filter_polygons_by_area(
+            polygons=polygons, min_area=None, max_area=maximum_detection_area
+        )
+    else:
+        polygons = filter_polygons_by_area(
+            polygons=polygons,
+            min_area=minimum_detection_area,
+            max_area=maximum_detection_area,
+        )
+    return [
+        approximate_polygon(polygon=polygon, percentage=approximation_percentage)
+        for polygon in polygons
+    ]
+
+
+def merge_class_lists(class_lists: List[List[str]]) -> List[str]:
+    unique_classes = set()
+
+    for class_list in class_lists:
+        for class_name in class_list:
+            unique_classes.add(class_name.lower())
+
+    return sorted(list(unique_classes))
+
+
+def build_class_index_mapping(
+    source_classes: List[str], target_classes: List[str]
+) -> Dict[int, int]:
+    index_mapping = {}
+
+    for i, class_name in enumerate(source_classes):
+        if class_name not in target_classes:
+            raise ValueError(
+                f"Class {class_name} not found in target classes. "
+                f"source_classes must be a subset of target_classes."
+            )
+        corresponding_index = target_classes.index(class_name)
+        index_mapping[i] = corresponding_index
+
+    return index_mapping
+
+
+def map_detections_class_id(
+    source_to_target_mapping: Dict[int, int], detections: Detections
+) -> Detections:
+    if detections.class_id is None:
+        raise ValueError("Detections must have class_id attribute.")
+    if set(np.unique(detections.class_id)) - set(source_to_target_mapping.keys()):
+        raise ValueError(
+            "Detections class_id must be a subset of source_to_target_mapping keys."
+        )
+
+    detections_copy = copy.deepcopy(detections)
+
+    if len(detections) > 0:
+        detections_copy.class_id = np.vectorize(source_to_target_mapping.get)(
+            detections_copy.class_id
+        )
+
+    return detections_copy
+
+
+def save_dataset_images(
+    images_directory_path: str, images: Dict[str, np.ndarray]
+) -> None:
+    Path(images_directory_path).mkdir(parents=True, exist_ok=True)
+
+    for image_name, image in images.items():
+        target_image_path = os.path.join(images_directory_path, image_name)
+        cv2.imwrite(target_image_path, image)
+
+
+def train_test_split(
+    data: List[T],
+    train_ratio: float = 0.8,
+    random_state: Optional[int] = None,
+    shuffle: bool = True,
+) -> Tuple[List[T], List[T]]:
+    """
+    Splits the data into two parts using the provided train_ratio.
+
+    Args:
+        data (List[T]): The data to split.
+        train_ratio (float): The ratio of the training set to the entire dataset.
+        random_state (Optional[int]): The seed for the random number generator.
+        shuffle (bool): Whether to shuffle the data before splitting.
+
+    Returns:
+        Tuple[List[T], List[T]]: The split data.
+    """
+    if random_state is not None:
+        random.seed(random_state)
+
+    if shuffle:
+        random.shuffle(data)
+
+    split_index = int(len(data) * train_ratio)
+    return data[:split_index], data[split_index:]

supervision/detection/annotate.py

@@ -47,13 +47,34 @@ class BoxAnnotator:
         """
         Draws bounding boxes on the frame using the detections provided.
 
-        Parameters:
+        Args:
             scene (np.ndarray): The image on which the bounding boxes will be drawn
             detections (Detections): The detections for which the bounding boxes will be drawn
-            labels (Optional[List[str]]): An optional list of labels corresponding to each detection. If labels is provided, the confidence score of the detection will be replaced with the label.
-            skip_label (bool): Is set to True, skips bounding box label annotation.
+            labels (Optional[List[str]]): An optional list of labels corresponding to each detection. If `labels` are not provided, corresponding `class_id` will be used as label.
+            skip_label (bool): Is set to `True`, skips bounding box label annotation.
         Returns:
             np.ndarray: The image with the bounding boxes drawn on it
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> classes = ['person', ...]
+            >>> image = ...
+            >>> detections = sv.Detections(...)
+
+            >>> box_annotator = sv.BoxAnnotator()
+            >>> labels = [
+            ...     f"{classes[class_id]} {confidence:0.2f}"
+            ...     for _, _, confidence, class_id, _
+            ...     in detections
+            ... ]
+            >>> annotated_frame = box_annotator.annotate(
+            ...     scene=image.copy(),
+            ...     detections=detections,
+            ...     labels=labels
+            ... )
+            ```
         """
         font = cv2.FONT_HERSHEY_SIMPLEX
         for i in range(len(detections)):
@@ -139,7 +160,7 @@ class MaskAnnotator:
         """
         Overlays the masks on the given image based on the provided detections, with a specified opacity.
 
-        Parameters:
+        Args:
             scene (np.ndarray): The image on which the masks will be overlaid
             detections (Detections): The detections for which the masks will be overlaid
             opacity (float): The opacity of the masks, between 0 and 1, default is 0.5
@@ -147,10 +168,10 @@ class MaskAnnotator:
         Returns:
             np.ndarray: The image with the masks overlaid
         """
-        for i in range(len(detections.xyxy)):
-            if detections.mask is None:
-                continue
+        if detections.mask is None:
+            return scene
 
+        for i in np.flip(np.argsort(detections.area)):
             class_id = (
                 detections.class_id[i] if detections.class_id is not None else None
             )

supervision/detection/core.py

@@ -1,12 +1,19 @@
 from __future__ import annotations
 
-from dataclasses import dataclass
+from dataclasses import astuple, dataclass
 from typing import Any, Iterator, List, Optional, Tuple, Union
 
+import cv2
 import numpy as np
 
-from supervision.detection.utils import non_max_suppression, xywh_to_xyxy
+from supervision.detection.utils import (
+    extract_ultralytics_masks,
+    non_max_suppression,
+    process_roboflow_result,
+    xywh_to_xyxy,
+)
 from supervision.geometry.core import Position
+from supervision.utils.internal import deprecated
 
 
 def _validate_xyxy(xyxy: Any, n: int) -> None:
@@ -20,7 +27,7 @@ def _validate_mask(mask: Any, n: int) -> None:
         isinstance(mask, np.ndarray) and len(mask.shape) == 3 and mask.shape[0] == n
     )
     if not is_valid:
-        raise ValueError("mask must be 3d np.ndarray with (n, W, H) shape")
+        raise ValueError("mask must be 3d np.ndarray with (n, H, W) shape")
 
 
 def _validate_class_id(class_id: Any, n: int) -> None:
@@ -51,19 +58,18 @@ def _validate_tracker_id(tracker_id: Any, n: int) -> None:
 class Detections:
     """
     Data class containing information about the detections in a video frame.
-
     Attributes:
         xyxy (np.ndarray): An array of shape `(n, 4)` containing the bounding boxes coordinates in format `[x1, y1, x2, y2]`
-        mask: (Optional[np.ndarray]): An array of shape `(n, W, H)` containing the segmentation masks.
-        class_id (Optional[np.ndarray]): An array of shape `(n,)` containing the class ids of the detections.
+        mask: (Optional[np.ndarray]): An array of shape `(n, H, W)` containing the segmentation masks.
         confidence (Optional[np.ndarray]): An array of shape `(n,)` containing the confidence scores of the detections.
+        class_id (Optional[np.ndarray]): An array of shape `(n,)` containing the class ids of the detections.
         tracker_id (Optional[np.ndarray]): An array of shape `(n,)` containing the tracker ids of the detections.
     """
 
     xyxy: np.ndarray
     mask: np.Optional[np.ndarray] = None
-    class_id: Optional[np.ndarray] = None
     confidence: Optional[np.ndarray] = None
+    class_id: Optional[np.ndarray] = None
     tracker_id: Optional[np.ndarray] = None
 
     def __post_init__(self):
@@ -137,7 +143,7 @@ class Detections:
     @classmethod
     def from_yolov5(cls, yolov5_results) -> Detections:
         """
-        Creates a Detections instance from a YOLOv5 output Detections
+        Creates a Detections instance from a [YOLOv5](https://github.com/ultralytics/yolov5) inference result.
 
         Args:
             yolov5_results (yolov5.models.common.Detections): The output Detections instance from YOLOv5
@@ -147,12 +153,14 @@ class Detections:
 
         Example:
             ```python
+            >>> import cv2
             >>> import torch
-            >>> from supervision import Detections
+            >>> import supervision as sv
 
+            >>> image = cv2.imread(SOURCE_IMAGE_PATH)
             >>> model = torch.hub.load('ultralytics/yolov5', 'yolov5s')
-            >>> results = model(IMAGE)
-            >>> detections = Detections.from_yolov5(results)
+            >>> result = model(image)
+            >>> detections = sv.Detections.from_yolov5(result)
             ```
         """
         yolov5_detections_predictions = yolov5_results.pred[0].cpu().cpu().numpy()
@@ -163,9 +171,12 @@ class Detections:
         )
 
     @classmethod
+    @deprecated(
+        "This method is deprecated and removed in 0.15.0 release. Use sv.Detections.from_ultralytics() instead."
+    )
     def from_yolov8(cls, yolov8_results) -> Detections:
         """
-        Creates a Detections instance from a YOLOv8 output Results
+        Creates a Detections instance from a [YOLOv8](https://github.com/ultralytics/ultralytics) inference result.
 
         Args:
             yolov8_results (ultralytics.yolo.engine.results.Results): The output Results instance from YOLOv8
@@ -175,24 +186,125 @@ class Detections:
 
         Example:
             ```python
+            >>> import cv2
             >>> from ultralytics import YOLO
-            >>> from supervision import Detections
+            >>> import supervision as sv
 
+            >>> image = cv2.imread(SOURCE_IMAGE_PATH)
             >>> model = YOLO('yolov8s.pt')
-            >>> yolov8_results = model(IMAGE)[0]
-            >>> detections = Detections.from_yolov8(yolov8_results)
+            >>> result = model(image)[0]
+            >>> detections = sv.Detections.from_yolov8(result)
             ```
         """
         return cls(
             xyxy=yolov8_results.boxes.xyxy.cpu().numpy(),
             confidence=yolov8_results.boxes.conf.cpu().numpy(),
             class_id=yolov8_results.boxes.cls.cpu().numpy().astype(int),
+            mask=extract_ultralytics_masks(yolov8_results),
+        )
+
+    @classmethod
+    def from_ultralytics(cls, ultralytics_results) -> Detections:
+        """
+        Creates a Detections instance from a [YOLOv8](https://github.com/ultralytics/ultralytics) inference result.
+
+        Args:
+            ultralytics_results (ultralytics.yolo.engine.results.Results): The output Results instance from YOLOv8
+
+        Returns:
+            Detections: A new Detections object.
+
+        Example:
+            ```python
+            >>> import cv2
+            >>> from ultralytics import YOLO, FastSAM, SAM, RTDETR
+            >>> import supervision as sv
+
+            >>> image = cv2.imread(SOURCE_IMAGE_PATH)
+            >>> model = YOLO('yolov8s.pt')
+            >>> model = SAM('sam_b.pt')
+            >>> model = SAM('mobile_sam.pt')
+            >>> model = FastSAM('FastSAM-s.pt')
+            >>> model = RTDETR('rtdetr-l.pt')
+            >>> # model inferences
+            >>> result = model(image)[0]
+            >>> # if tracker is enabled
+            >>> result = model.track(image)[0]
+            >>> detections = sv.Detections.from_ultralytics(result)
+            ```
+        """
+        return cls(
+            xyxy=ultralytics_results.boxes.xyxy.cpu().numpy(),
+            confidence=ultralytics_results.boxes.conf.cpu().numpy(),
+            class_id=ultralytics_results.boxes.cls.cpu().numpy().astype(int),
+            mask=extract_ultralytics_masks(ultralytics_results),
+            tracker_id=ultralytics_results.boxes.id.int().cpu().numpy()
+            if ultralytics_results.boxes.id is not None
+            else None,
+        )
+
+    @classmethod
+    def from_yolo_nas(cls, yolo_nas_results) -> Detections:
+        """
+        Creates a Detections instance from a [YOLO-NAS](https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md) inference result.
+
+        Args:
+            yolo_nas_results (super_gradients.training.models.prediction_results.ImageDetectionPrediction): The output Results instance from YOLO-NAS
+
+        Returns:
+            Detections: A new Detections object.
+
+        Example:
+            ```python
+            >>> import cv2
+            >>> from super_gradients.training import models
+            >>> import supervision as sv
+
+            >>> image = cv2.imread(SOURCE_IMAGE_PATH)
+            >>> model = models.get('yolo_nas_l', pretrained_weights="coco")
+            >>> result = list(model.predict(image, conf=0.35))[0]
+            >>> detections = sv.Detections.from_yolo_nas(result)
+            ```
+        """
+        return cls(
+            xyxy=yolo_nas_results.prediction.bboxes_xyxy,
+            confidence=yolo_nas_results.prediction.confidence,
+            class_id=yolo_nas_results.prediction.labels.astype(int),
+        )
+
+    @classmethod
+    def from_mmdetection(cls, mmdet_results) -> Detections:
+        """
+        Creates a Detections instance from a [mmdetection](https://github.com/open-mmlab/mmdetection) inference result.
+        Also supported for [mmyolo](https://github.com/open-mmlab/mmyolo)
+
+        Args:
+            mmdet_results (mmdet.structures.DetDataSample): The output Results instance from MMDetection
+
+        Returns:
+            Detections: A new Detections object.
+
+        Example:
+            ```python
+            >>> import cv2
+            >>> import supervision as sv
+            >>> from mmdet.apis import DetInferencer
+
+            >>> inferencer = DetInferencer(model_name, checkpoint, device)
+            >>> mmdet_result = inferencer(SOURCE_IMAGE_PATH, out_dir='./output', return_datasample=True)["predictions"][0]
+            >>> detections = sv.Detections.from_mmdet(mmdet_result)
+            ```
+        """
+        return cls(
+            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),
         )
 
     @classmethod
     def from_transformers(cls, transformers_results: dict) -> Detections:
         """
-        Creates a Detections instance from Object Detection Transformer output Results
+        Creates a Detections instance from object detection [transformer](https://github.com/huggingface/transformers) inference result.
 
         Returns:
             Detections: A new Detections object.
@@ -205,6 +317,32 @@ class Detections:
 
     @classmethod
     def from_detectron2(cls, detectron2_results) -> Detections:
+        """
+        Create a Detections object from the [Detectron2](https://github.com/facebookresearch/detectron2) inference result.
+
+        Args:
+            detectron2_results: The output of a Detectron2 model containing instances with prediction data.
+
+        Returns:
+            (Detections): A Detections object containing the bounding boxes, class IDs, and confidences of the predictions.
+
+        Example:
+            ```python
+            >>> import cv2
+            >>> from detectron2.engine import DefaultPredictor
+            >>> from detectron2.config import get_cfg
+            >>> import supervision as sv
+
+            >>> image = cv2.imread(SOURCE_IMAGE_PATH)
+            >>> cfg = get_cfg()
+            >>> cfg.merge_from_file("path/to/config.yaml")
+            >>> cfg.MODEL.WEIGHTS = "path/to/model_weights.pth"
+            >>> predictor = DefaultPredictor(cfg)
+            >>> result = predictor(image)
+
+            >>> detections = sv.Detections.from_detectron2(result)
+            ```
+        """
         return cls(
             xyxy=detectron2_results["instances"].pred_boxes.tensor.cpu().numpy(),
             confidence=detectron2_results["instances"].scores.cpu().numpy(),
@@ -216,33 +354,52 @@ class Detections:
 
     @classmethod
     def from_roboflow(cls, roboflow_result: dict, class_list: List[str]) -> Detections:
-        xyxy = []
-        confidence = []
-        class_id = []
+        """
+        Create a Detections object from the [Roboflow](https://roboflow.com/) API inference result.
 
-        for prediction in roboflow_result["predictions"]:
-            x = prediction["x"]
-            y = prediction["y"]
-            width = prediction["width"]
-            height = prediction["height"]
-            x_min = x - width / 2
-            y_min = y - height / 2
-            x_max = x_min + width
-            y_max = y_min + height
-            xyxy.append([x_min, y_min, x_max, y_max])
-            class_id.append(class_list.index(prediction["class"]))
-            confidence.append(prediction["confidence"])
+        Args:
+            roboflow_result (dict): The result from the Roboflow API containing predictions.
+            class_list (List[str]): A list of class names corresponding to the class IDs in the API result.
 
+        Returns:
+            (Detections): A Detections object containing the bounding boxes, class IDs, and confidences of the predictions.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> roboflow_result = {
+            ...     "predictions": [
+            ...         {
+            ...             "x": 0.5,
+            ...             "y": 0.5,
+            ...             "width": 0.2,
+            ...             "height": 0.3,
+            ...             "class": "person",
+            ...             "confidence": 0.9
+            ...         },
+            ...         # ... more predictions ...
+            ...     ]
+            ... }
+            >>> class_list = ["person", "car", "dog"]
+
+            >>> detections = sv.Detections.from_roboflow(roboflow_result, class_list)
+            ```
+        """
+        xyxy, confidence, class_id, masks = process_roboflow_result(
+            roboflow_result=roboflow_result, class_list=class_list
+        )
         return Detections(
-            xyxy=np.array(xyxy),
-            confidence=np.array(confidence),
-            class_id=np.array(class_id).astype(int),
+            xyxy=xyxy,
+            confidence=confidence,
+            class_id=class_id,
+            mask=masks,
         )
 
     @classmethod
     def from_sam(cls, sam_result: List[dict]) -> Detections:
         """
-        Creates a Detections instance from Segment Anything Model (SAM) by Meta AI.
+        Creates a Detections instance from [Segment Anything Model](https://github.com/facebookresearch/segment-anything) inference result.
 
         Args:
             sam_result (List[dict]): The output Results instance from SAM
@@ -252,8 +409,8 @@ class Detections:
 
         Example:
             ```python
-            >>> from segment_anything import sam_model_registry, SamAutomaticMaskGenerator
             >>> import supervision as sv
+            >>> from segment_anything import sam_model_registry, SamAutomaticMaskGenerator
 
             >>> sam = sam_model_registry[MODEL_TYPE](checkpoint=CHECKPOINT_PATH).to(device=DEVICE)
             >>> mask_generator = SamAutomaticMaskGenerator(sam)
@@ -271,24 +428,111 @@ class Detections:
         return Detections(xyxy=xywh_to_xyxy(boxes_xywh=xywh), mask=mask)
 
     @classmethod
-    def from_coco_annotations(cls, coco_annotation: dict) -> Detections:
-        xyxy, class_id = [], []
+    def from_paddledet(cls, paddledet_result) -> Detections:
+        """
+        Creates a Detections instance from [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection) inference result.
 
-        for annotation in coco_annotation:
-            x_min, y_min, width, height = annotation["bbox"]
-            xyxy.append([x_min, y_min, x_min + width, y_min + height])
-            class_id.append(annotation["category_id"])
+        Args:
+            paddledet_result (List[dict]): The output Results instance from SAM
 
-        return cls(xyxy=np.array(xyxy), class_id=np.array(class_id))
+        Returns:
+            Detections: A new Detections object.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+            >>> import paddle
+            >>> from ppdet.engine import Trainer
+            >>> from ppdet.core.workspace import load_config
+
+            >>> weights = (...)
+            >>> config = (...)
+
+            >>> cfg = load_config(config)
+            >>> trainer = Trainer(cfg, mode='test')
+            >>> trainer.load_weights(weights)
+
+            >>> paddledet_result = trainer.predict([images])[0]
+
+            >>> detections = sv.Detections.from_paddledet(paddledet_result=paddledet_result)
+            ```
+        """
+        return cls(
+            xyxy=paddledet_result["bbox"][:, 2:6],
+            confidence=paddledet_result["bbox"][:, 1],
+            class_id=paddledet_result["bbox"][:, 0].astype(int),
+        )
 
     @classmethod
     def empty(cls) -> Detections:
+        """
+        Create an empty Detections object with no bounding boxes, confidences, or class IDs.
+
+        Returns:
+            (Detections): An empty Detections object.
+
+        Example:
+            ```python
+            >>> from supervision import Detections
+
+            >>> empty_detections = Detections.empty()
+            ```
+        """
         return cls(
             xyxy=np.empty((0, 4), dtype=np.float32),
             confidence=np.array([], dtype=np.float32),
             class_id=np.array([], dtype=int),
         )
 
+    @classmethod
+    def merge(cls, detections_list: List[Detections]) -> Detections:
+        """
+        Merge a list of Detections objects into a single Detections object.
+
+        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`.
+
+        Args:
+            detections_list (List[Detections]): A list of Detections objects to merge.
+
+        Returns:
+            (Detections): A single Detections object containing the merged data from the input list.
+
+        Example:
+            ```python
+            >>> from supervision import Detections
+
+            >>> detections_1 = Detections(...)
+            >>> detections_2 = Detections(...)
+
+            >>> merged_detections = Detections.merge([detections_1, detections_2])
+            ```
+        """
+        if len(detections_list) == 0:
+            return Detections.empty()
+
+        detections_tuples_list = [astuple(detection) for detection in detections_list]
+        xyxy, mask, confidence, class_id, tracker_id = [
+            list(field) for field in zip(*detections_tuples_list)
+        ]
+
+        all_not_none = lambda l: all(x is not None for x in l)
+
+        xyxy = np.vstack(xyxy)
+        mask = np.vstack(mask) if all_not_none(mask) else None
+        confidence = np.hstack(confidence) if all_not_none(confidence) else None
+        class_id = np.hstack(class_id) if all_not_none(class_id) else None
+        tracker_id = np.hstack(tracker_id) if all_not_none(tracker_id) else None
+
+        return cls(
+            xyxy=xyxy,
+            mask=mask,
+            confidence=confidence,
+            class_id=class_id,
+            tracker_id=tracker_id,
+        )
+
     def get_anchor_coordinates(self, anchor: Position) -> np.ndarray:
         """
         Returns the bounding box coordinates for a specific anchor.
@@ -313,22 +557,43 @@ class Detections:
 
         raise ValueError(f"{anchor} is not supported.")
 
-    def __getitem__(self, index: np.ndarray) -> Detections:
-        if isinstance(index, np.ndarray) and (
-            index.dtype == bool or index.dtype == int
-        ):
-            return Detections(
-                xyxy=self.xyxy[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,
-                tracker_id=self.tracker_id[index]
-                if self.tracker_id is not None
-                else None,
-            )
-        raise TypeError(
-            f"Detections.__getitem__ not supported for index of type {type(index)}."
+    def __getitem__(
+        self, index: Union[int, slice, List[int], np.ndarray]
+    ) -> Detections:
+        """
+        Get a subset of the Detections object.
+
+        Args:
+            index (Union[int, slice, List[int], np.ndarray]): The index or indices of the subset of the Detections
+
+        Returns:
+            (Detections): A subset of the Detections object.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> detections = sv.Detections(...)
+
+            >>> first_detection = detections[0]
+
+            >>> first_10_detections = detections[0:10]
+
+            >>> some_detections = detections[[0, 2, 4]]
+
+            >>> class_0_detections = detections[detections.class_id == 0]
+
+            >>> high_confidence_detections = detections[detections.confidence > 0.5]
+            ```
+        """
+        if isinstance(index, int):
+            index = [index]
+        return Detections(
+            xyxy=self.xyxy[index],
+            mask=self.mask[index] if self.mask is not None else None,
+            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,
+            tracker_id=self.tracker_id[index] if self.tracker_id is not None else None,
         )
 
     @property
@@ -341,7 +606,7 @@ class Detections:
           np.ndarray: An array of floats containing the area of each detection in the format of `(area_1, area_2, ..., area_n)`, where n is the number of detections.
         """
         if self.mask is not None:
-            return np.ndarray([np.sum(mask) for mask in self.mask])
+            return np.array([np.sum(mask) for mask in self.mask])
         else:
             return self.box_area
 

supervision/detection/line_counter.py

@@ -1,4 +1,4 @@
-from typing import Dict
+from typing import Dict, Optional
 
 import cv2
 import numpy as np
@@ -35,7 +35,7 @@ class LineZone:
             detections (Detections): The detections for which to update the counts.
 
         """
-        for xyxy, confidence, class_id, tracker_id in detections:
+        for xyxy, _, confidence, class_id, tracker_id in detections:
             # handle detections with no tracker_id
             if tracker_id is None:
                 continue
@@ -81,6 +81,8 @@ class LineZoneAnnotator:
         text_scale: float = 0.5,
         text_offset: float = 1.5,
         text_padding: int = 10,
+        custom_in_text: Optional[str] = None,
+        custom_out_text: Optional[str] = None,
     ):
         """
         Initialize the LineCounterAnnotator object with default values.
@@ -102,6 +104,8 @@ class LineZoneAnnotator:
         self.text_scale: float = text_scale
         self.text_offset: float = text_offset
         self.text_padding: int = text_padding
+        self.custom_in_text: str = custom_in_text
+        self.custom_out_text: str = custom_out_text
 
     def annotate(self, frame: np.ndarray, line_counter: LineZone) -> np.ndarray:
         """
@@ -141,8 +145,16 @@ class LineZoneAnnotator:
             lineType=cv2.LINE_AA,
         )
 
-        in_text = f"in: {line_counter.in_count}"
-        out_text = f"out: {line_counter.out_count}"
+        in_text = (
+            f"{self.custom_in_text}: {line_counter.in_count}"
+            if self.custom_in_text is not None
+            else f"in: {line_counter.in_count}"
+        )
+        out_text = (
+            f"{self.custom_out_text}: {line_counter.out_count}"
+            if self.custom_out_text is not None
+            else f"out: {line_counter.out_count}"
+        )
 
         (in_text_width, in_text_height), _ = cv2.getTextSize(
             in_text, cv2.FONT_HERSHEY_SIMPLEX, self.text_scale, self.text_thickness
@@ -219,3 +231,4 @@ class LineZoneAnnotator:
             self.text_thickness,
             cv2.LINE_AA,
         )
+        return frame

supervision/detection/tools/polygon_zone.py

@@ -5,7 +5,7 @@ import cv2
 import numpy as np
 
 from supervision import Detections
-from supervision.detection.utils import clip_boxes, generate_2d_mask
+from supervision.detection.utils import clip_boxes, polygon_to_mask
 from supervision.draw.color import Color
 from supervision.draw.utils import draw_polygon, draw_text
 from supervision.geometry.core import Position
@@ -17,7 +17,7 @@ class PolygonZone:
     A class for defining a polygon-shaped zone within a frame for detecting objects.
 
     Attributes:
-        polygon (np.ndarray): A numpy array defining the polygon vertices
+        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_position (Position): The position within the bounding box that triggers the zone (default: Position.BOTTOM_CENTER)
         current_count (int): The current count of detected objects within the zone
@@ -30,13 +30,13 @@ class PolygonZone:
         frame_resolution_wh: Tuple[int, int],
         triggering_position: Position = Position.BOTTOM_CENTER,
     ):
-        self.polygon = polygon
+        self.polygon = polygon.astype(int)
         self.frame_resolution_wh = frame_resolution_wh
         self.triggering_position = triggering_position
         self.current_count = 0
 
         width, height = frame_resolution_wh
-        self.mask = generate_2d_mask(
+        self.mask = polygon_to_mask(
             polygon=polygon, resolution_wh=(width + 1, height + 1)
         )
 

supervision/detection/utils.py

@@ -1,11 +1,13 @@
-from typing import Tuple
+from typing import List, Optional, Tuple
 
 import cv2
 import numpy as np
 
+MIN_POLYGON_POINT_COUNT = 3
 
-def generate_2d_mask(polygon: np.ndarray, resolution_wh: Tuple[int, int]) -> np.ndarray:
-    """Generate a 2D mask from a polygon.
+
+def polygon_to_mask(polygon: np.ndarray, resolution_wh: Tuple[int, int]) -> np.ndarray:
+    """Generate a mask from a polygon.
 
     Args:
         polygon (np.ndarray): The polygon for which the mask should be generated, given as a list of vertices.
@@ -15,7 +17,7 @@ def generate_2d_mask(polygon: np.ndarray, resolution_wh: Tuple[int, int]) -> np.
         np.ndarray: The generated 2D mask, where the polygon is marked with `1`'s and the rest is filled with `0`'s.
     """
     width, height = resolution_wh
-    mask = np.zeros((height, width), dtype=np.uint8)
+    mask = np.zeros((height, width))
     cv2.fillPoly(mask, [polygon], color=1)
     return mask
 
@@ -146,3 +148,196 @@ def mask_to_xyxy(masks: np.ndarray) -> np.ndarray:
             bboxes[i, :] = [x_min, y_min, x_max, y_max]
 
     return bboxes
+
+
+def mask_to_polygons(mask: np.ndarray) -> List[np.ndarray]:
+    """
+    Converts a binary mask to a list of polygons.
+
+    Parameters:
+        mask (np.ndarray): A binary mask represented as a 2D NumPy array of shape `(H, W)`,
+            where H and W are the height and width of the mask, respectively.
+
+    Returns:
+        List[np.ndarray]: A list of polygons, where each polygon is represented by a NumPy array of shape `(N, 2)`,
+            containing the `x`, `y` coordinates of the points. Polygons with fewer points than `MIN_POLYGON_POINT_COUNT = 3`
+            are excluded from the output.
+    """
+
+    contours, _ = cv2.findContours(
+        mask.astype(np.uint8), cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE
+    )
+    return [
+        np.squeeze(contour, axis=1)
+        for contour in contours
+        if contour.shape[0] >= MIN_POLYGON_POINT_COUNT
+    ]
+
+
+def filter_polygons_by_area(
+    polygons: List[np.ndarray],
+    min_area: Optional[float] = None,
+    max_area: Optional[float] = None,
+) -> List[np.ndarray]:
+    """
+    Filters a list of polygons based on their area.
+
+    Parameters:
+        polygons (List[np.ndarray]): A list of polygons, where each polygon is represented by a NumPy array of shape `(N, 2)`,
+            containing the `x`, `y` coordinates of the points.
+        min_area (Optional[float]): The minimum area threshold. Only polygons with an area greater than or equal to this value
+            will be included in the output. If set to None, no minimum area constraint will be applied.
+        max_area (Optional[float]): The maximum area threshold. Only polygons with an area less than or equal to this value
+            will be included in the output. If set to None, no maximum area constraint will be applied.
+
+    Returns:
+        List[np.ndarray]: A new list of polygons containing only those with areas within the specified thresholds.
+    """
+    if min_area is None and max_area is None:
+        return polygons
+    ares = [cv2.contourArea(polygon) for polygon in polygons]
+    return [
+        polygon
+        for polygon, area in zip(polygons, ares)
+        if (min_area is None or area >= min_area)
+        and (max_area is None or area <= max_area)
+    ]
+
+
+def polygon_to_xyxy(polygon: np.ndarray) -> np.ndarray:
+    """
+    Converts a polygon represented by a NumPy array into a bounding box.
+
+    Parameters:
+        polygon (np.ndarray): A polygon represented by a NumPy array of shape `(N, 2)`,
+            containing the `x`, `y` coordinates of the points.
+
+    Returns:
+        np.ndarray: A 1D NumPy array containing the bounding box `(x_min, y_min, x_max, y_max)` of the input polygon.
+    """
+    x_min, y_min = np.min(polygon, axis=0)
+    x_max, y_max = np.max(polygon, axis=0)
+    return np.array([x_min, y_min, x_max, y_max])
+
+
+def approximate_polygon(
+    polygon: np.ndarray, percentage: float, epsilon_step: float = 0.05
+) -> np.ndarray:
+    """
+    Approximates a given polygon by reducing a certain percentage of points.
+
+    This function uses the Ramer-Douglas-Peucker algorithm to simplify the input polygon by reducing the number of points
+    while preserving the general shape.
+
+    Parameters:
+        polygon (np.ndarray): A 2D NumPy array of shape `(N, 2)` containing the `x`, `y` coordinates of the input polygon's points.
+        percentage (float): The percentage of points to be removed from the input polygon, in the range `[0, 1)`.
+        epsilon_step (float): Approximation accuracy step. Epsilon is the maximum distance between the original curve and its approximation.
+
+    Returns:
+        np.ndarray: A new 2D NumPy array of shape `(M, 2)`, where `M <= N * (1 - percentage)`, containing the `x`, `y` coordinates of the
+            approximated polygon's points.
+    """
+
+    if percentage < 0 or percentage >= 1:
+        raise ValueError("Percentage must be in the range [0, 1).")
+
+    target_points = max(int(len(polygon) * (1 - percentage)), 3)
+
+    if len(polygon) <= target_points:
+        return polygon
+
+    epsilon = 0
+    approximated_points = polygon
+    while True:
+        epsilon += epsilon_step
+        new_approximated_points = cv2.approxPolyDP(polygon, epsilon, closed=True)
+        if len(new_approximated_points) > target_points:
+            approximated_points = new_approximated_points
+        else:
+            break
+
+    return np.squeeze(approximated_points, axis=1)
+
+
+def extract_ultralytics_masks(yolov8_results) -> Optional[np.ndarray]:
+    if not yolov8_results.masks:
+        return None
+
+    orig_shape = yolov8_results.orig_shape
+    inference_shape = tuple(yolov8_results.masks.data.shape[1:])
+
+    gain = 0
+    pad = (0, 0)
+
+    if inference_shape != orig_shape:
+        gain = min(
+            inference_shape[0] / orig_shape[0],
+            inference_shape[1] / orig_shape[1],
+        )
+        pad = (
+            (inference_shape[1] - orig_shape[1] * gain) / 2,
+            (inference_shape[0] - orig_shape[0] * gain) / 2,
+        )
+
+    top, left = int(pad[1]), int(pad[0])
+    bottom, right = int(inference_shape[0] - pad[1]), int(inference_shape[1] - pad[0])
+
+    mask_maps = []
+    masks = yolov8_results.masks.data.cpu().numpy()
+    for i in range(masks.shape[0]):
+        mask = masks[i]
+        mask = mask[top:bottom, left:right]
+
+        if mask.shape != orig_shape:
+            mask = cv2.resize(mask, (orig_shape[1], orig_shape[0]))
+
+        mask_maps.append(mask)
+
+    return np.asarray(mask_maps, dtype=bool)
+
+
+def process_roboflow_result(
+    roboflow_result: dict, class_list: List[str]
+) -> Tuple[np.ndarray, np.ndarray, np.ndarray, Optional[np.ndarray]]:
+    if not roboflow_result["predictions"]:
+        return np.empty((0, 4)), np.empty(0), np.empty(0), None
+
+    xyxy = []
+    confidence = []
+    class_id = []
+    masks = []
+
+    image_width = int(roboflow_result["image"]["width"])
+    image_height = int(roboflow_result["image"]["height"])
+
+    for prediction in roboflow_result["predictions"]:
+        x = prediction["x"]
+        y = prediction["y"]
+        width = prediction["width"]
+        height = prediction["height"]
+        x_min = x - width / 2
+        y_min = y - height / 2
+        x_max = x_min + width
+        y_max = y_min + height
+
+        xyxy.append([x_min, y_min, x_max, y_max])
+        class_id.append(class_list.index(prediction["class"]))
+        confidence.append(prediction["confidence"])
+
+        if "points" not in prediction:
+            continue
+
+        polygon = np.array(
+            [[point["x"], point["y"]] for point in prediction["points"]], dtype=int
+        )
+
+        mask = polygon_to_mask(polygon, resolution_wh=(image_width, image_height))
+        masks.append(mask)
+
+    xyxy = np.array(xyxy)
+    confidence = np.array(confidence)
+    class_id = np.array(class_id).astype(int)
+    masks = np.array(masks, dtype=bool) if len(masks) > 0 else None
+
+    return xyxy, confidence, class_id, masks

supervision/metrics/detection.py

@@ -0,0 +1,792 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Callable, List, Optional, Tuple
+
+import matplotlib
+import matplotlib.pyplot as plt
+import numpy as np
+
+from supervision.dataset.core import DetectionDataset
+from supervision.detection.core import Detections
+from supervision.detection.utils import box_iou_batch
+
+
+def detections_to_tensor(
+    detections: Detections, with_confidence: bool = False
+) -> np.ndarray:
+    """
+    Convert Supervision Detections to numpy tensors for further computation
+    Args:
+        detections (sv.Detections): Detections/Targets in the format of sv.Detections
+        with_confidence (bool): Whether to include confidence in the tensor
+    Returns:
+        (np.ndarray): Detections as numpy tensors as in (xyxy, class_id, confidence) order
+    """
+    if detections.class_id is None:
+        raise ValueError(
+            "ConfusionMatrix can only be calculated for Detections with class_id"
+        )
+
+    arrays_to_concat = [detections.xyxy, np.expand_dims(detections.class_id, 1)]
+
+    if with_confidence:
+        if detections.confidence is None:
+            raise ValueError(
+                "ConfusionMatrix can only be calculated for Detections with confidence"
+            )
+        arrays_to_concat.append(np.expand_dims(detections.confidence, 1))
+
+    return np.concatenate(arrays_to_concat, axis=1)
+
+
+def validate_input_tensors(predictions: List[np.ndarray], targets: List[np.ndarray]):
+    """
+    Checks for shape consistency of input tensors.
+    """
+    if len(predictions) != len(targets):
+        raise ValueError(
+            f"Number of predictions ({len(predictions)}) and targets ({len(targets)}) must be equal."
+        )
+    if len(predictions) > 0:
+        if not isinstance(predictions[0], np.ndarray) or not isinstance(
+            targets[0], np.ndarray
+        ):
+            raise ValueError(
+                f"Predictions and targets must be lists of numpy arrays. Got {type(predictions[0])} and {type(targets[0])} instead."
+            )
+        if predictions[0].shape[1] != 6:
+            raise ValueError(
+                f"Predictions must have shape (N, 6). Got {predictions[0].shape} instead."
+            )
+        if targets[0].shape[1] != 5:
+            raise ValueError(
+                f"Targets must have shape (N, 5). Got {targets[0].shape} instead."
+            )
+
+
+@dataclass
+class ConfusionMatrix:
+    """
+    Confusion matrix for object detection tasks.
+
+    Attributes:
+        matrix (np.ndarray): An 2D `np.ndarray` of shape `(len(classes) + 1, len(classes) + 1)` containing the number of `TP`, `FP`, `FN` and `TN` for each class.
+        classes (List[str]): Model class names.
+        conf_threshold (float): Detection confidence threshold between `0` and `1`. Detections with lower confidence will be excluded from the matrix.
+        iou_threshold (float): Detection IoU threshold between `0` and `1`. Detections with lower IoU will be classified as `FP`.
+    """
+
+    matrix: np.ndarray
+    classes: List[str]
+    conf_threshold: float
+    iou_threshold: float
+
+    @classmethod
+    def from_detections(
+        cls,
+        predictions: List[Detections],
+        targets: List[Detections],
+        classes: List[str],
+        conf_threshold: float = 0.3,
+        iou_threshold: float = 0.5,
+    ) -> ConfusionMatrix:
+        """
+        Calculate confusion matrix based on predicted and ground-truth detections.
+
+        Args:
+            targets (List[Detections]): Detections objects from ground-truth.
+            predictions (List[Detections]): Detections objects predicted by the model.
+            classes (List[str]): Model class names.
+            conf_threshold (float): Detection confidence threshold between `0` and `1`. Detections with lower confidence will be excluded.
+            iou_threshold (float): Detection IoU threshold between `0` and `1`. Detections with lower IoU will be classified as `FP`.
+
+        Returns:
+            ConfusionMatrix: New instance of ConfusionMatrix.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> targets = [
+            ...     sv.Detections(...),
+            ...     sv.Detections(...)
+            ... ]
+
+            >>> predictions = [
+            ...     sv.Detections(...),
+            ...     sv.Detections(...)
+            ... ]
+
+            >>> confusion_matrix = sv.ConfusionMatrix.from_detections(
+            ...     predictions=predictions,
+            ...     targets=target,
+            ...     classes=['person', ...]
+            ... )
+
+            >>> confusion_matrix.matrix
+            array([
+                [0., 0., 0., 0.],
+                [0., 1., 0., 1.],
+                [0., 1., 1., 0.],
+                [1., 1., 0., 0.]
+            ])
+            ```
+        """
+
+        prediction_tensors = []
+        target_tensors = []
+        for prediction, target in zip(predictions, targets):
+            prediction_tensors.append(
+                detections_to_tensor(prediction, with_confidence=True)
+            )
+            target_tensors.append(detections_to_tensor(target, with_confidence=False))
+        return cls.from_tensors(
+            predictions=prediction_tensors,
+            targets=target_tensors,
+            classes=classes,
+            conf_threshold=conf_threshold,
+            iou_threshold=iou_threshold,
+        )
+
+    @classmethod
+    def from_tensors(
+        cls,
+        predictions: List[np.ndarray],
+        targets: List[np.ndarray],
+        classes: List[str],
+        conf_threshold: float = 0.3,
+        iou_threshold: float = 0.5,
+    ) -> ConfusionMatrix:
+        """
+        Calculate confusion matrix based on predicted and ground-truth detections.
+
+        Args:
+            predictions (List[np.ndarray]): Each element of the list describes a single image and has `shape = (M, 6)` where `M` is the number of detected objects. Each row is expected to be in `(x_min, y_min, x_max, y_max, class, conf)` format.
+            targets (List[np.ndarray]): Each element of the list describes a single image and has `shape = (N, 5)` where `N` is the number of ground-truth objects. Each row is expected to be in `(x_min, y_min, x_max, y_max, class)` format.
+            classes (List[str]): Model class names.
+            conf_threshold (float): Detection confidence threshold between `0` and `1`. Detections with lower confidence will be excluded.
+            iou_threshold (float): Detection iou  threshold between `0` and `1`. Detections with lower iou will be classified as `FP`.
+
+        Returns:
+            ConfusionMatrix: New instance of ConfusionMatrix.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> targets = (
+            ...     [
+            ...         array(
+            ...             [
+            ...                 [0.0, 0.0, 3.0, 3.0, 1],
+            ...                 [2.0, 2.0, 5.0, 5.0, 1],
+            ...                 [6.0, 1.0, 8.0, 3.0, 2],
+            ...             ]
+            ...         ),
+            ...         array([1.0, 1.0, 2.0, 2.0, 2]),
+            ...     ]
+            ... )
+
+            >>> predictions = [
+            ...     array(
+            ...         [
+            ...             [0.0, 0.0, 3.0, 3.0, 1, 0.9],
+            ...             [0.1, 0.1, 3.0, 3.0, 0, 0.9],
+            ...             [6.0, 1.0, 8.0, 3.0, 1, 0.8],
+            ...             [1.0, 6.0, 2.0, 7.0, 1, 0.8],
+            ...         ]
+            ...     ),
+            ...     array([[1.0, 1.0, 2.0, 2.0, 2, 0.8]])
+            ... ]
+
+            >>> confusion_matrix = sv.ConfusionMatrix.from_tensors(
+            ...     predictions=predictions,
+            ...     targets=targets,
+            ...     classes=['person', ...]
+            ... )
+
+            >>> confusion_matrix.matrix
+            array([
+                [0., 0., 0., 0.],
+                [0., 1., 0., 1.],
+                [0., 1., 1., 0.],
+                [1., 1., 0., 0.]
+            ])
+            ```
+        """
+        validate_input_tensors(predictions, targets)
+
+        num_classes = len(classes)
+        matrix = np.zeros((num_classes + 1, num_classes + 1))
+        for true_batch, detection_batch in zip(targets, predictions):
+            matrix += cls.evaluate_detection_batch(
+                predictions=detection_batch,
+                targets=true_batch,
+                num_classes=num_classes,
+                conf_threshold=conf_threshold,
+                iou_threshold=iou_threshold,
+            )
+        return cls(
+            matrix=matrix,
+            classes=classes,
+            conf_threshold=conf_threshold,
+            iou_threshold=iou_threshold,
+        )
+
+    @staticmethod
+    def evaluate_detection_batch(
+        predictions: np.ndarray,
+        targets: np.ndarray,
+        num_classes: int,
+        conf_threshold: float,
+        iou_threshold: float,
+    ) -> np.ndarray:
+        """
+        Calculate confusion matrix for a batch of detections for a single image.
+
+        Args:
+            predictions (np.ndarray): Batch prediction. Describes a single image and has `shape = (M, 6)` where `M` is the number of detected objects. Each row is expected to be in `(x_min, y_min, x_max, y_max, class, conf)` format.
+            targets (np.ndarray): Batch target labels. Describes a single image and has `shape = (N, 5)` where `N` is the number of ground-truth objects. Each row is expected to be in `(x_min, y_min, x_max, y_max, class)` format.
+            num_classes (int): Number of classes.
+            conf_threshold (float): Detection confidence threshold between `0` and `1`. Detections with lower confidence will be excluded.
+            iou_threshold (float): Detection iou  threshold between `0` and `1`. Detections with lower iou will be classified as `FP`.
+
+        Returns:
+            np.ndarray: Confusion matrix based on a single image.
+        """
+        result_matrix = np.zeros((num_classes + 1, num_classes + 1))
+
+        conf_idx = 5
+        confidence = predictions[:, conf_idx]
+        detection_batch_filtered = predictions[confidence > conf_threshold]
+
+        class_id_idx = 4
+        true_classes = np.array(targets[:, class_id_idx], dtype=np.int16)
+        detection_classes = np.array(
+            detection_batch_filtered[:, class_id_idx], dtype=np.int16
+        )
+        true_boxes = targets[:, :class_id_idx]
+        detection_boxes = detection_batch_filtered[:, :class_id_idx]
+
+        iou_batch = box_iou_batch(
+            boxes_true=true_boxes, boxes_detection=detection_boxes
+        )
+        matched_idx = np.asarray(iou_batch > iou_threshold).nonzero()
+
+        if matched_idx[0].shape[0]:
+            matches = np.stack(
+                (matched_idx[0], matched_idx[1], iou_batch[matched_idx]), axis=1
+            )
+            matches = ConfusionMatrix._drop_extra_matches(matches=matches)
+        else:
+            matches = np.zeros((0, 3))
+
+        matched_true_idx, matched_detection_idx, _ = matches.transpose().astype(
+            np.int16
+        )
+
+        for i, true_class_value in enumerate(true_classes):
+            j = matched_true_idx == i
+            if matches.shape[0] > 0 and sum(j) == 1:
+                result_matrix[
+                    true_class_value, detection_classes[matched_detection_idx[j]]
+                ] += 1  # TP
+            else:
+                result_matrix[true_class_value, num_classes] += 1  # FN
+
+        for i, detection_class_value in enumerate(detection_classes):
+            if not any(matched_detection_idx == i):
+                result_matrix[num_classes, detection_class_value] += 1  # FP
+
+        return result_matrix
+
+    @staticmethod
+    def _drop_extra_matches(matches: np.ndarray) -> np.ndarray:
+        """
+        Deduplicate matches. If there are multiple matches for the same true or predicted box,
+        only the one with the highest IoU is kept.
+        """
+        if matches.shape[0] > 0:
+            matches = matches[matches[:, 2].argsort()[::-1]]
+            matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
+            matches = matches[matches[:, 2].argsort()[::-1]]
+            matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
+        return matches
+
+    @classmethod
+    def benchmark(
+        cls,
+        dataset: DetectionDataset,
+        callback: Callable[[np.ndarray], Detections],
+        conf_threshold: float = 0.3,
+        iou_threshold: float = 0.5,
+    ) -> ConfusionMatrix:
+        """
+        Calculate confusion matrix from dataset and callback function.
+
+        Args:
+            dataset (DetectionDataset): Object detection dataset used for evaluation.
+            callback (Callable[[np.ndarray], Detections]): Function that takes an image as input and returns Detections object.
+            conf_threshold (float): Detection confidence threshold between `0` and `1`. Detections with lower confidence will be excluded.
+            iou_threshold (float): Detection IoU threshold between `0` and `1`. Detections with lower IoU will be classified as `FP`.
+
+        Returns:
+            ConfusionMatrix: New instance of ConfusionMatrix.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+            >>> from ultralytics import YOLO
+
+            >>> dataset = sv.DetectionDataset.from_yolo(...)
+
+            >>> model = YOLO(...)
+            >>> def callback(image: np.ndarray) -> sv.Detections:
+            ...     result = model(image)[0]
+            ...     return sv.Detections.from_yolov8(result)
+
+            >>> confusion_matrix = sv.ConfusionMatrix.benchmark(
+            ...     dataset = dataset,
+            ...     callback = callback
+            ... )
+
+            >>> confusion_matrix.matrix
+            array([
+                [0., 0., 0., 0.],
+                [0., 1., 0., 1.],
+                [0., 1., 1., 0.],
+                [1., 1., 0., 0.]
+            ])
+            ```
+        """
+        predictions, targets = [], []
+        for img_name, img in dataset.images.items():
+            predictions_batch = callback(img)
+            predictions.append(predictions_batch)
+            targets_batch = dataset.annotations[img_name]
+            targets.append(targets_batch)
+        return cls.from_detections(
+            predictions=predictions,
+            targets=targets,
+            classes=dataset.classes,
+            conf_threshold=conf_threshold,
+            iou_threshold=iou_threshold,
+        )
+
+    def plot(
+        self,
+        save_path: Optional[str] = None,
+        title: Optional[str] = None,
+        classes: Optional[List[str]] = None,
+        normalize: bool = False,
+        fig_size: Tuple[int, int] = (12, 10),
+    ) -> matplotlib.figure.Figure:
+        """
+        Create confusion matrix plot and save it at selected location.
+
+        Args:
+            save_path (Optional[str]): Path to save the plot. If not provided, plot will be displayed.
+            title (Optional[str]): Title of the plot.
+            classes (Optional[List[str]]): List of classes to be displayed on the plot. If not provided, all classes will be displayed.
+            normalize (bool): If True, normalize the confusion matrix.
+            fig_size (Tuple[int, int]): Size of the plot.
+
+        Returns:
+            matplotlib.figure.Figure: Confusion matrix plot.
+        """
+
+        array = self.matrix.copy()
+
+        if normalize:
+            eps = 1e-8
+            array = array / (array.sum(0).reshape(1, -1) + eps)
+
+        array[array < 0.005] = np.nan
+
+        fig, ax = plt.subplots(figsize=fig_size, tight_layout=True, facecolor="white")
+
+        class_names = classes if classes is not None else self.classes
+        use_labels_for_ticks = class_names is not None and (0 < len(class_names) < 99)
+        if use_labels_for_ticks:
+            x_tick_labels = class_names + ["FN"]
+            y_tick_labels = class_names + ["FP"]
+            num_ticks = len(x_tick_labels)
+        else:
+            x_tick_labels = None
+            y_tick_labels = None
+            num_ticks = len(array)
+        im = ax.imshow(array, cmap="Blues")
+
+        cbar = ax.figure.colorbar(im, ax=ax)
+        cbar.mappable.set_clim(vmin=0, vmax=np.nanmax(array))
+
+        if x_tick_labels is None:
+            tick_interval = 2
+        else:
+            tick_interval = 1
+        ax.set_xticks(np.arange(0, num_ticks, tick_interval), labels=x_tick_labels)
+        ax.set_yticks(np.arange(0, num_ticks, tick_interval), labels=y_tick_labels)
+
+        plt.setp(ax.get_xticklabels(), rotation=90, ha="right", rotation_mode="default")
+
+        labelsize = 10 if num_ticks < 50 else 8
+        ax.tick_params(axis="both", which="both", labelsize=labelsize)
+
+        if num_ticks < 30:
+            for i in range(array.shape[0]):
+                for j in range(array.shape[1]):
+                    n_preds = array[i, j]
+                    if not np.isnan(n_preds):
+                        ax.text(
+                            j,
+                            i,
+                            f"{n_preds:.2f}" if normalize else f"{n_preds:.0f}",
+                            ha="center",
+                            va="center",
+                            color="black"
+                            if n_preds < 0.5 * np.nanmax(array)
+                            else "white",
+                        )
+
+        if title:
+            ax.set_title(title, fontsize=20)
+
+        ax.set_xlabel("Predicted")
+        ax.set_ylabel("True")
+        ax.set_facecolor("white")
+        if save_path:
+            fig.savefig(
+                save_path, dpi=250, facecolor=fig.get_facecolor(), transparent=True
+            )
+        return fig
+
+
+@dataclass(frozen=True)
+class MeanAveragePrecision:
+    """
+    Mean Average Precision for object detection tasks.
+
+    Attributes:
+        map50_95 (float): Mean Average Precision (mAP) calculated over IoU thresholds ranging from `0.50` to `0.95` with a step size of `0.05`.
+        map50 (float): Mean Average Precision (mAP) calculated specifically at an IoU threshold of `0.50`.
+        map75 (float): Mean Average Precision (mAP) calculated specifically at an IoU threshold of `0.75`.
+        per_class_ap50_95 (np.ndarray): Average Precision (AP) values calculated over IoU thresholds ranging from `0.50` to `0.95` with a step size of `0.05`, provided for each individual class.
+    """
+
+    map50_95: float
+    map50: float
+    map75: float
+    per_class_ap50_95: np.ndarray
+
+    @classmethod
+    def from_detections(
+        cls,
+        predictions: List[Detections],
+        targets: List[Detections],
+    ) -> MeanAveragePrecision:
+        """
+        Calculate mean average precision based on predicted and ground-truth detections.
+
+        Args:
+            targets (List[Detections]): Detections objects from ground-truth.
+            predictions (List[Detections]): Detections objects predicted by the model.
+        Returns:
+            MeanAveragePrecision: New instance of ConfusionMatrix.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> targets = [
+            ...     sv.Detections(...),
+            ...     sv.Detections(...)
+            ... ]
+
+            >>> predictions = [
+            ...     sv.Detections(...),
+            ...     sv.Detections(...)
+            ... ]
+
+            >>> mean_average_precision = sv.MeanAveragePrecision.from_detections(
+            ...     predictions=predictions,
+            ...     targets=target,
+            ... )
+
+            >>> mean_average_precison.map50_95
+            0.2899
+            ```
+        """
+        prediction_tensors = []
+        target_tensors = []
+        for prediction, target in zip(predictions, targets):
+            prediction_tensors.append(
+                detections_to_tensor(prediction, with_confidence=True)
+            )
+            target_tensors.append(detections_to_tensor(target, with_confidence=False))
+        return cls.from_tensors(
+            predictions=prediction_tensors,
+            targets=target_tensors,
+        )
+
+    @classmethod
+    def benchmark(
+        cls,
+        dataset: DetectionDataset,
+        callback: Callable[[np.ndarray], Detections],
+    ) -> MeanAveragePrecision:
+        """
+        Calculate mean average precision from dataset and callback function.
+
+        Args:
+            dataset (DetectionDataset): Object detection dataset used for evaluation.
+            callback (Callable[[np.ndarray], Detections]): Function that takes an image as input and returns Detections object.
+        Returns:
+            MeanAveragePrecision: New instance of MeanAveragePrecision.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+            >>> from ultralytics import YOLO
+
+            >>> dataset = sv.DetectionDataset.from_yolo(...)
+
+            >>> model = YOLO(...)
+            >>> def callback(image: np.ndarray) -> sv.Detections:
+            ...     result = model(image)[0]
+            ...     return sv.Detections.from_yolov8(result)
+
+            >>> mean_average_precision = sv.MeanAveragePrecision.benchmark(
+            ...     dataset = dataset,
+            ...     callback = callback
+            ... )
+
+            >>> mean_average_precision.map50_95
+            0.433
+            ```
+        """
+        predictions, targets = [], []
+        for img_name, img in dataset.images.items():
+            predictions_batch = callback(img)
+            predictions.append(predictions_batch)
+            targets_batch = dataset.annotations[img_name]
+            targets.append(targets_batch)
+        return cls.from_detections(
+            predictions=predictions,
+            targets=targets,
+        )
+
+    @classmethod
+    def from_tensors(
+        cls,
+        predictions: List[np.ndarray],
+        targets: List[np.ndarray],
+    ) -> MeanAveragePrecision:
+        """
+        Calculate Mean Average Precision based on predicted and ground-truth detections at different threshold.
+
+        Args:
+            predictions (List[np.ndarray]): Each element of the list describes a single image and has `shape = (M, 6)` where `M` is the number of detected objects. Each row is expected to be in `(x_min, y_min, x_max, y_max, class, conf)` format.
+            targets (List[np.ndarray]): Each element of the list describes a single image and has `shape = (N, 5)` where `N` is the number of ground-truth objects. Each row is expected to be in `(x_min, y_min, x_max, y_max, class)` format.
+        Returns:
+            MeanAveragePrecision: New instance of MeanAveragePrecision.
+
+        Example:
+            ```python
+            >>> import supervision as sv
+
+            >>> targets = (
+            ...     [
+            ...         array(
+            ...             [
+            ...                 [0.0, 0.0, 3.0, 3.0, 1],
+            ...                 [2.0, 2.0, 5.0, 5.0, 1],
+            ...                 [6.0, 1.0, 8.0, 3.0, 2],
+            ...             ]
+            ...         ),
+            ...         array([1.0, 1.0, 2.0, 2.0, 2]),
+            ...     ]
+            ... )
+
+            >>> predictions = [
+            ...     array(
+            ...         [
+            ...             [0.0, 0.0, 3.0, 3.0, 1, 0.9],
+            ...             [0.1, 0.1, 3.0, 3.0, 0, 0.9],
+            ...             [6.0, 1.0, 8.0, 3.0, 1, 0.8],
+            ...             [1.0, 6.0, 2.0, 7.0, 1, 0.8],
+            ...         ]
+            ...     ),
+            ...     array([[1.0, 1.0, 2.0, 2.0, 2, 0.8]])
+            ... ]
+
+            >>> mean_average_precison = sv.MeanAveragePrecision.from_tensors(
+            ...     predictions=predictions,
+            ...     targets=targets,
+            ... )
+
+            >>> mean_average_precison.map50_95
+            0.2899
+            ```
+        """
+        validate_input_tensors(predictions, targets)
+        iou_thresholds = np.linspace(0.5, 0.95, 10)
+        stats = []
+
+        # Gather matching stats for predictions and targets
+        for true_objs, predicted_objs in zip(targets, predictions):
+            if predicted_objs.shape[0] == 0:
+                if true_objs.shape[0]:
+                    stats.append(
+                        (
+                            np.zeros((0, iou_thresholds.size), dtype=bool),
+                            *np.zeros((2, 0)),
+                            true_objs[:, 4],
+                        )
+                    )
+                continue
+
+            if true_objs.shape[0]:
+                matches = cls._match_detection_batch(
+                    predicted_objs, true_objs, iou_thresholds
+                )
+                stats.append(
+                    (
+                        matches,
+                        predicted_objs[:, 5],
+                        predicted_objs[:, 4],
+                        true_objs[:, 4],
+                    )
+                )
+
+        # Compute average precisions if any matches exist
+        if stats:
+            concatenated_stats = [np.concatenate(items, 0) for items in zip(*stats)]
+            average_precisions = cls._average_precisions_per_class(*concatenated_stats)
+            map50 = average_precisions[:, 0].mean()
+            map75 = average_precisions[:, 5].mean()
+            map50_95 = average_precisions.mean()
+        else:
+            map50, map75, map50_95 = 0, 0, 0
+            average_precisions = []
+
+        return cls(
+            map50_95=map50_95,
+            map50=map50,
+            map75=map75,
+            per_class_ap50_95=average_precisions,
+        )
+
+    @staticmethod
+    def compute_average_precision(recall: np.ndarray, precision: np.ndarray) -> float:
+        """
+        Compute the average precision using 101-point interpolation (COCO), given the recall and precision curves.
+
+        Args:
+            recall (np.ndarray): The recall curve.
+            precision (np.ndarray): The precision curve.
+
+        Returns:
+            float: Average precision.
+        """
+        extended_recall = np.concatenate(([0.0], recall, [1.0]))
+        extended_precision = np.concatenate(([1.0], precision, [0.0]))
+        max_accumulated_precision = np.flip(
+            np.maximum.accumulate(np.flip(extended_precision))
+        )
+        interpolated_recall_levels = np.linspace(0, 1, 101)
+        interpolated_precision = np.interp(
+            interpolated_recall_levels, extended_recall, max_accumulated_precision
+        )
+        average_precision = np.trapz(interpolated_precision, interpolated_recall_levels)
+        return average_precision
+
+    @staticmethod
+    def _match_detection_batch(
+        predictions: np.ndarray, targets: np.ndarray, iou_thresholds: np.ndarray
+    ) -> np.ndarray:
+        """
+        Match predictions with target labels based on IoU levels.
+
+        Args:
+            predictions (np.ndarray): Batch prediction. Describes a single image and has `shape = (M, 6)` where `M` is the number of detected objects. Each row is expected to be in `(x_min, y_min, x_max, y_max, class, conf)` format.
+            targets (np.ndarray): Batch target labels. Describes a single image and has `shape = (N, 5)` where `N` is the number of ground-truth objects. Each row is expected to be in `(x_min, y_min, x_max, y_max, class)` format.
+            iou_thresholds (np.ndarray): Array contains different IoU thresholds.
+
+        Returns:
+            np.ndarray: Matched prediction with target labels result.
+        """
+        num_predictions, num_iou_levels = predictions.shape[0], iou_thresholds.shape[0]
+        correct = np.zeros((num_predictions, num_iou_levels), dtype=bool)
+        iou = box_iou_batch(targets[:, :4], predictions[:, :4])
+        correct_class = targets[:, 4:5] == predictions[:, 4]
+
+        for i, iou_level in enumerate(iou_thresholds):
+            matched_indices = np.where((iou >= iou_level) & correct_class)
+
+            if matched_indices[0].shape[0]:
+                combined_indices = np.stack(matched_indices, axis=1)
+                iou_values = iou[matched_indices][:, None]
+                matches = np.hstack([combined_indices, iou_values])
+
+                if matched_indices[0].shape[0] > 1:
+                    matches = matches[matches[:, 2].argsort()[::-1]]
+                    matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
+                    matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
+
+                correct[matches[:, 1].astype(int), i] = True
+
+        return correct
+
+    @staticmethod
+    def _average_precisions_per_class(
+        matches: np.ndarray,
+        prediction_confidence: np.ndarray,
+        prediction_class_ids: np.ndarray,
+        true_class_ids: np.ndarray,
+        eps: float = 1e-16,
+    ) -> np.ndarray:
+        """
+        Compute the average precision, given the recall and precision curves.
+        Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
+
+        Args:
+            matches (np.ndarray): True positives.
+            prediction_confidence (np.ndarray): Objectness value from 0-1.
+            prediction_class_ids (np.ndarray): Predicted object classes.
+            true_class_ids (np.ndarray): True object classes.
+            eps (float, optional): Small value to prevent division by zero.
+
+        Returns:
+            np.ndarray: Average precision for different IoU levels.
+        """
+        sorted_indices = np.argsort(-prediction_confidence)
+        matches = matches[sorted_indices]
+        prediction_class_ids = prediction_class_ids[sorted_indices]
+
+        unique_classes, class_counts = np.unique(true_class_ids, return_counts=True)
+        num_classes = unique_classes.shape[0]
+
+        average_precisions = np.zeros((num_classes, matches.shape[1]))
+
+        for class_idx, class_id in enumerate(unique_classes):
+            is_class = prediction_class_ids == class_id
+            total_true = class_counts[class_idx]
+            total_prediction = is_class.sum()
+
+            if total_prediction == 0 or total_true == 0:
+                continue
+
+            false_positives = (1 - matches[is_class]).cumsum(0)
+            true_positives = matches[is_class].cumsum(0)
+            recall = true_positives / (total_true + eps)
+            precision = true_positives / (true_positives + false_positives)
+
+            for iou_level_idx in range(matches.shape[1]):
+                average_precisions[
+                    class_idx, iou_level_idx
+                ] = MeanAveragePrecision.compute_average_precision(
+                    recall[:, iou_level_idx], precision[:, iou_level_idx]
+                )
+
+        return average_precisions

supervision/tracker/byte_tracker/basetrack.py

@@ -0,0 +1,55 @@
+from collections import OrderedDict
+from enum import Enum
+
+import numpy as np
+
+
+class TrackState(Enum):
+    New = 0
+    Tracked = 1
+    Lost = 2
+    Removed = 3
+
+
+class BaseTrack:
+    _count = 0
+
+    def __init__(self):
+        self.track_id = 0
+        self.is_activated = False
+        self.state = TrackState.New
+
+        self.history = OrderedDict()
+        self.features = []
+        self.curr_feature = None
+        self.score = 0
+        self.start_frame = 0
+        self.frame_id = 0
+        self.time_since_update = 0
+
+        # multi-camera
+        self.location = (np.inf, np.inf)
+
+    @property
+    def end_frame(self) -> int:
+        return self.frame_id
+
+    @staticmethod
+    def next_id() -> int:
+        BaseTrack._count += 1
+        return BaseTrack._count
+
+    def activate(self, *args):
+        raise NotImplementedError
+
+    def predict(self):
+        raise NotImplementedError
+
+    def update(self, *args, **kwargs):
+        raise NotImplementedError
+
+    def mark_lost(self):
+        self.state = TrackState.Lost
+
+    def mark_removed(self):
+        self.state = TrackState.Removed

supervision/tracker/byte_tracker/core.py

@@ -0,0 +1,466 @@
+from typing import List, Tuple
+
+import numpy as np
+
+from supervision.detection.core import Detections
+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
+
+
+class STrack(BaseTrack):
+    shared_kalman = KalmanFilter()
+
+    def __init__(self, tlwh, score, class_ids):
+        # wait activate
+        self._tlwh = np.asarray(tlwh, dtype=np.float32)
+        self.kalman_filter = None
+        self.mean, self.covariance = None, None
+        self.is_activated = False
+
+        self.score = score
+        self.class_ids = class_ids
+        self.tracklet_len = 0
+
+    def predict(self):
+        mean_state = self.mean.copy()
+        if self.state != TrackState.Tracked:
+            mean_state[7] = 0
+        self.mean, self.covariance = self.kalman_filter.predict(
+            mean_state, self.covariance
+        )
+
+    @staticmethod
+    def multi_predict(stracks):
+        if len(stracks) > 0:
+            multi_mean = np.asarray([st.mean.copy() for st in stracks])
+            multi_covariance = np.asarray([st.covariance for st in stracks])
+            for i, st in enumerate(stracks):
+                if st.state != TrackState.Tracked:
+                    multi_mean[i][7] = 0
+            multi_mean, multi_covariance = STrack.shared_kalman.multi_predict(
+                multi_mean, multi_covariance
+            )
+            for i, (mean, cov) in enumerate(zip(multi_mean, multi_covariance)):
+                stracks[i].mean = mean
+                stracks[i].covariance = cov
+
+    def activate(self, kalman_filter, frame_id):
+        """Start a new tracklet"""
+        self.kalman_filter = kalman_filter
+        self.track_id = self.next_id()
+        self.mean, self.covariance = self.kalman_filter.initiate(
+            self.tlwh_to_xyah(self._tlwh)
+        )
+
+        self.tracklet_len = 0
+        self.state = TrackState.Tracked
+        if frame_id == 1:
+            self.is_activated = True
+        self.frame_id = frame_id
+        self.start_frame = frame_id
+
+    def re_activate(self, new_track, frame_id, new_id=False):
+        self.mean, self.covariance = self.kalman_filter.update(
+            self.mean, self.covariance, self.tlwh_to_xyah(new_track.tlwh)
+        )
+        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.score = new_track.score
+
+    def update(self, new_track, frame_id):
+        """
+        Update a matched track
+        :type new_track: STrack
+        :type frame_id: int
+        :type update_feature: bool
+        :return:
+        """
+        self.frame_id = frame_id
+        self.tracklet_len += 1
+
+        new_tlwh = new_track.tlwh
+        self.mean, self.covariance = self.kalman_filter.update(
+            self.mean, self.covariance, self.tlwh_to_xyah(new_tlwh)
+        )
+        self.state = TrackState.Tracked
+        self.is_activated = True
+
+        self.score = new_track.score
+
+    @property
+    def tlwh(self):
+        """Get current position in bounding box format `(top left x, top left y,
+        width, height)`.
+        """
+        if self.mean is None:
+            return self._tlwh.copy()
+        ret = self.mean[:4].copy()
+        ret[2] *= ret[3]
+        ret[:2] -= ret[2:] / 2
+        return ret
+
+    @property
+    def tlbr(self):
+        """Convert bounding box to format `(min x, min y, max x, max y)`, i.e.,
+        `(top left, bottom right)`.
+        """
+        ret = self.tlwh.copy()
+        ret[2:] += ret[:2]
+        return ret
+
+    @staticmethod
+    def tlwh_to_xyah(tlwh):
+        """Convert bounding box to format `(center x, center y, aspect ratio,
+        height)`, where the aspect ratio is `width / height`.
+        """
+        ret = np.asarray(tlwh).copy()
+        ret[:2] += ret[2:] / 2
+        ret[2] /= ret[3]
+        return ret
+
+    def to_xyah(self):
+        return self.tlwh_to_xyah(self.tlwh)
+
+    @staticmethod
+    def tlbr_to_tlwh(tlbr):
+        ret = np.asarray(tlbr).copy()
+        ret[2:] -= ret[:2]
+        return ret
+
+    @staticmethod
+    def tlwh_to_tlbr(tlwh):
+        ret = np.asarray(tlwh).copy()
+        ret[2:] += ret[:2]
+        return ret
+
+    def __repr__(self):
+        return "OT_{}_({}-{})".format(self.track_id, self.start_frame, self.end_frame)
+
+
+def detections2boxes(detections: Detections) -> np.ndarray:
+    """
+    Convert Supervision Detections to numpy tensors for further computation.
+    Args:
+        detections (Detections): Detections/Targets in the format of sv.Detections.
+    Returns:
+        (np.ndarray): Detections as numpy tensors as in `(x_min, y_min, x_max, y_max, confidence, class_id)` order.
+    """
+    return np.hstack(
+        (
+            detections.xyxy,
+            detections.confidence[:, np.newaxis],
+            detections.class_id[:, np.newaxis],
+        )
+    )
+
+
+class ByteTrack:
+    """
+    Initialize the ByteTrack object.
+
+    Parameters:
+        track_thresh (float, optional): Detection confidence threshold for track activation.
+        track_buffer (int, optional): Number of frames to buffer when a track is lost.
+        match_thresh (float, optional): Threshold for matching tracks with detections.
+        frame_rate (int, optional): The frame rate of the video.
+    """
+
+    def __init__(
+        self,
+        track_thresh: float = 0.25,
+        track_buffer: int = 30,
+        match_thresh: float = 0.8,
+        frame_rate: int = 30,
+    ):
+        self.track_thresh = track_thresh
+        self.match_thresh = match_thresh
+
+        self.frame_id = 0
+        self.det_thresh = self.track_thresh + 0.1
+        self.max_time_lost = int(frame_rate / 30.0 * track_buffer)
+        self.kalman_filter = KalmanFilter()
+
+        self.tracked_tracks: List[STrack] = []
+        self.lost_tracks: List[STrack] = []
+        self.removed_tracks: List[STrack] = []
+
+    def update_with_detections(self, detections: Detections) -> Detections:
+        """
+        Updates the tracker with the provided detections and returns the updated detection results.
+
+        Parameters:
+            detections: The new detections to update with.
+        Returns:
+            Detection: The updated detection results that now include tracking IDs.
+        Example:
+            ```python
+            >>> import supervision as sv
+            >>> from ultralytics import YOLO
+
+            >>> model = YOLO(...)
+            >>> byte_tracker = sv.ByteTrack()
+            >>> annotator = sv.BoxAnnotator()
+
+            >>> def callback(frame: np.ndarray, index: int) -> np.ndarray:
+            ...     results = model(frame)[0]
+            ...     detections = sv.Detections.from_yolov8(results)
+            ...     detections = byte_tracker.update_from_detections(detections=detections)
+            ...     labels = [
+            ...         f"#{tracker_id} {model.model.names[class_id]} {confidence:0.2f}"
+            ...         for _, _, confidence, class_id, tracker_id
+            ...         in detections
+            ...     ]
+            ...     return annotator.annotate(scene=frame.copy(), detections=detections, labels=labels)
+
+            >>> sv.process_video(
+            ...     source_path='...',
+            ...     target_path='...',
+            ...     callback=callback
+            ... )
+            ```
+        """
+
+        tracks = self.update_with_tensors(
+            tensors=detections2boxes(detections=detections)
+        )
+        detections = Detections.empty()
+        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
+            )
+
+        return detections
+
+    def update_with_tensors(self, tensors: np.ndarray) -> List[STrack]:
+        """
+        Updates the tracker with the provided tensors and returns the updated tracks.
+
+        Parameters:
+            tensors: The new tensors to update with.
+
+        Returns:
+            List[STrack]: Updated tracks.
+        """
+        self.frame_id += 1
+        activated_starcks = []
+        refind_stracks = []
+        lost_stracks = []
+        removed_stracks = []
+
+        class_ids = tensors[:, 5]
+        scores = tensors[:, 4]
+        bboxes = tensors[:, :4]
+
+        remain_inds = scores > self.track_thresh
+        inds_low = scores > 0.1
+        inds_high = scores < self.track_thresh
+
+        inds_second = np.logical_and(inds_low, inds_high)
+        dets_second = bboxes[inds_second]
+        dets = bboxes[remain_inds]
+        scores_keep = scores[remain_inds]
+        scores_second = scores[inds_second]
+
+        class_ids_keep = class_ids[remain_inds]
+        class_ids_second = class_ids[inds_second]
+
+        if len(dets) > 0:
+            """Detections"""
+            detections = [
+                STrack(STrack.tlbr_to_tlwh(tlbr), s, c)
+                for (tlbr, s, c) in zip(dets, scores_keep, class_ids_keep)
+            ]
+        else:
+            detections = []
+
+        """ Add newly detected tracklets to tracked_stracks"""
+        unconfirmed = []
+        tracked_stracks = []  # type: list[STrack]
+        for track in self.tracked_tracks:
+            if not track.is_activated:
+                unconfirmed.append(track)
+            else:
+                tracked_stracks.append(track)
+
+        """ Step 2: First association, with high score detection boxes"""
+        strack_pool = joint_tracks(tracked_stracks, self.lost_tracks)
+        # Predict the current location with KF
+        STrack.multi_predict(strack_pool)
+        dists = matching.iou_distance(strack_pool, detections)
+
+        dists = matching.fuse_score(dists, detections)
+        matches, u_track, u_detection = matching.linear_assignment(
+            dists, thresh=self.match_thresh
+        )
+
+        for itracked, idet in matches:
+            track = strack_pool[itracked]
+            det = detections[idet]
+            if track.state == TrackState.Tracked:
+                track.update(detections[idet], self.frame_id)
+                activated_starcks.append(track)
+            else:
+                track.re_activate(det, self.frame_id, new_id=False)
+                refind_stracks.append(track)
+
+        """ Step 3: Second association, with low score detection boxes"""
+        # association the untrack to the low score detections
+        if len(dets_second) > 0:
+            """Detections"""
+            detections_second = [
+                STrack(STrack.tlbr_to_tlwh(tlbr), s, c)
+                for (tlbr, s, c) in zip(dets_second, scores_second, class_ids_second)
+            ]
+        else:
+            detections_second = []
+        r_tracked_stracks = [
+            strack_pool[i]
+            for i in u_track
+            if strack_pool[i].state == TrackState.Tracked
+        ]
+        dists = matching.iou_distance(r_tracked_stracks, detections_second)
+        matches, u_track, u_detection_second = matching.linear_assignment(
+            dists, thresh=0.5
+        )
+        for itracked, idet in matches:
+            track = r_tracked_stracks[itracked]
+            det = detections_second[idet]
+            if track.state == TrackState.Tracked:
+                track.update(det, self.frame_id)
+                activated_starcks.append(track)
+            else:
+                track.re_activate(det, self.frame_id, new_id=False)
+                refind_stracks.append(track)
+
+        for it in u_track:
+            track = r_tracked_stracks[it]
+            if not track.state == TrackState.Lost:
+                track.mark_lost()
+                lost_stracks.append(track)
+
+        """Deal with unconfirmed tracks, usually tracks with only one beginning frame"""
+        detections = [detections[i] for i in u_detection]
+        dists = matching.iou_distance(unconfirmed, detections)
+
+        dists = matching.fuse_score(dists, detections)
+        matches, u_unconfirmed, u_detection = matching.linear_assignment(
+            dists, thresh=0.7
+        )
+        for itracked, idet in matches:
+            unconfirmed[itracked].update(detections[idet], self.frame_id)
+            activated_starcks.append(unconfirmed[itracked])
+        for it in u_unconfirmed:
+            track = unconfirmed[it]
+            track.mark_removed()
+            removed_stracks.append(track)
+
+        """ Step 4: Init new stracks"""
+        for inew in u_detection:
+            track = detections[inew]
+            if track.score < self.det_thresh:
+                continue
+            track.activate(self.kalman_filter, self.frame_id)
+            activated_starcks.append(track)
+        """ Step 5: Update state"""
+        for track in self.lost_tracks:
+            if self.frame_id - track.end_frame > self.max_time_lost:
+                track.mark_removed()
+                removed_stracks.append(track)
+
+        self.tracked_tracks = [
+            t for t in self.tracked_tracks if t.state == TrackState.Tracked
+        ]
+        self.tracked_tracks = joint_tracks(self.tracked_tracks, activated_starcks)
+        self.tracked_tracks = joint_tracks(self.tracked_tracks, refind_stracks)
+        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.tracked_tracks, self.lost_tracks = remove_duplicate_tracks(
+            self.tracked_tracks, self.lost_tracks
+        )
+        output_stracks = [track for track in self.tracked_tracks if track.is_activated]
+
+        return output_stracks
+
+
+def joint_tracks(
+    track_list_a: List[STrack], track_list_b: List[STrack]
+) -> List[STrack]:
+    """
+    Joins two lists of tracks, ensuring that the resulting list does not
+    contain tracks with duplicate 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).
+
+    Returns:
+        Combined list of tracks from track_list_a and track_list_b without duplicate 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)
+            result.append(track)
+
+    return result
+
+
+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.
+
+    Parameters:
+        track_list_a: List of tracks (with track_id attribute).
+        track_list_b: List of tracks (with 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}
+
+    for track_id in track_ids_b:
+        tracks.pop(track_id, None)
+
+    return list(tracks.values())
+
+
+def remove_duplicate_tracks(tracks_a: List, tracks_b: List) -> Tuple[List, List]:
+    pairwise_distance = matching.iou_distance(tracks_a, tracks_b)
+    matching_pairs = np.where(pairwise_distance < 0.15)
+
+    duplicates_a, duplicates_b = set(), set()
+    for track_index_a, track_index_b in zip(*matching_pairs):
+        time_a = tracks_a[track_index_a].frame_id - tracks_a[track_index_a].start_frame
+        time_b = tracks_b[track_index_b].frame_id - tracks_b[track_index_b].start_frame
+        if time_a > time_b:
+            duplicates_b.add(track_index_b)
+        else:
+            duplicates_a.add(track_index_a)
+
+    result_a = [
+        track for index, track in enumerate(tracks_a) if index not in duplicates_a
+    ]
+    result_b = [
+        track for index, track in enumerate(tracks_b) if index not in duplicates_b
+    ]
+
+    return result_a, result_b

supervision/tracker/byte_tracker/kalman_filter.py

@@ -0,0 +1,190 @@
+from typing import Tuple
+
+import numpy as np
+import scipy.linalg
+
+
+class KalmanFilter:
+    """
+    A simple Kalman filter for tracking bounding boxes in image space.
+
+    The 8-dimensional state space
+
+        x, y, a, h, vx, vy, va, vh
+
+    contains the bounding box center position (x, y), aspect ratio a, height h,
+    and their respective velocities.
+
+    Object motion follows a constant velocity model. The bounding box location
+    (x, y, a, h) is taken as direct observation of the state space (linear
+    observation model).
+    """
+
+    def __init__(self):
+        ndim, dt = 4, 1.0
+
+        self._motion_mat = np.eye(2 * ndim, 2 * ndim)
+        for i in range(ndim):
+            self._motion_mat[i, ndim + i] = dt
+        self._update_mat = np.eye(ndim, 2 * ndim)
+        self._std_weight_position = 1.0 / 20
+        self._std_weight_velocity = 1.0 / 160
+
+    def initiate(self, measurement: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
+        """
+        Create track from an unassociated measurement.
+
+        Args:
+            measurement (ndarray): Bounding box coordinates (x, y, a, h) with center position (x, y), aspect ratio a, and height h.
+
+        Returns:
+            Tuple[ndarray, ndarray]: Returns the mean vector (8 dimensional) and covariance matrix (8x8 dimensional) of the new track. Unobserved velocities are initialized to 0 mean.
+        """
+        mean_pos = measurement
+        mean_vel = np.zeros_like(mean_pos)
+        mean = np.r_[mean_pos, mean_vel]
+
+        std = [
+            2 * self._std_weight_position * measurement[3],
+            2 * self._std_weight_position * measurement[3],
+            1e-2,
+            2 * self._std_weight_position * measurement[3],
+            10 * self._std_weight_velocity * measurement[3],
+            10 * self._std_weight_velocity * measurement[3],
+            1e-5,
+            10 * self._std_weight_velocity * measurement[3],
+        ]
+        covariance = np.diag(np.square(std))
+        return mean, covariance
+
+    def predict(
+        self, mean: np.ndarray, covariance: np.ndarray
+    ) -> Tuple[np.ndarray, np.ndarray]:
+        """
+        Run Kalman filter prediction step.
+
+        Args:
+            mean (ndarray): The 8 dimensional mean vector of the object state at the previous time step.
+            covariance (ndarray): The 8x8 dimensional covariance matrix of the object state at the previous time step.
+
+        Returns:
+            Tuple[ndarray, ndarray]: Returns the mean vector and covariance matrix of the predicted state. Unobserved velocities are initialized to 0 mean.
+        """
+        std_pos = [
+            self._std_weight_position * mean[3],
+            self._std_weight_position * mean[3],
+            1e-2,
+            self._std_weight_position * mean[3],
+        ]
+        std_vel = [
+            self._std_weight_velocity * mean[3],
+            self._std_weight_velocity * mean[3],
+            1e-5,
+            self._std_weight_velocity * mean[3],
+        ]
+        motion_cov = np.diag(np.square(np.r_[std_pos, std_vel]))
+
+        mean = np.dot(mean, self._motion_mat.T)
+        covariance = (
+            np.linalg.multi_dot((self._motion_mat, covariance, self._motion_mat.T))
+            + motion_cov
+        )
+
+        return mean, covariance
+
+    def project(
+        self, mean: np.ndarray, covariance: np.ndarray
+    ) -> Tuple[np.ndarray, np.ndarray]:
+        """
+        Project state distribution to measurement space.
+
+        Args:
+            mean (ndarray): The state's mean vector (8 dimensional array).
+            covariance (ndarray): The state's covariance matrix (8x8 dimensional).
+
+        Returns:
+            Tuple[ndarray, ndarray]: Returns the projected mean and covariance matrix of the given state estimate.
+        """
+        std = [
+            self._std_weight_position * mean[3],
+            self._std_weight_position * mean[3],
+            1e-1,
+            self._std_weight_position * mean[3],
+        ]
+        innovation_cov = np.diag(np.square(std))
+
+        mean = np.dot(self._update_mat, mean)
+        covariance = np.linalg.multi_dot(
+            (self._update_mat, covariance, self._update_mat.T)
+        )
+        return mean, covariance + innovation_cov
+
+    def multi_predict(
+        self, mean: np.ndarray, covariance: np.ndarray
+    ) -> Tuple[np.ndarray, np.ndarray]:
+        """
+        Run Kalman filter prediction step (Vectorized version).
+
+        Args:
+            mean (ndarray): The Nx8 dimensional mean matrix of the object states at the previous time step.
+            covariance (ndarray): The Nx8x8 dimensional covariance matrices of the object states at the previous time step.
+
+        Returns:
+            Tuple[ndarray, ndarray]: Returns the mean vector and covariance matrix of the predicted state. Unobserved velocities are initialized to 0 mean.
+        """
+        std_pos = [
+            self._std_weight_position * mean[:, 3],
+            self._std_weight_position * mean[:, 3],
+            1e-2 * np.ones_like(mean[:, 3]),
+            self._std_weight_position * mean[:, 3],
+        ]
+        std_vel = [
+            self._std_weight_velocity * mean[:, 3],
+            self._std_weight_velocity * mean[:, 3],
+            1e-5 * np.ones_like(mean[:, 3]),
+            self._std_weight_velocity * mean[:, 3],
+        ]
+        sqr = np.square(np.r_[std_pos, std_vel]).T
+
+        motion_cov = []
+        for i in range(len(mean)):
+            motion_cov.append(np.diag(sqr[i]))
+        motion_cov = np.asarray(motion_cov)
+
+        mean = np.dot(mean, self._motion_mat.T)
+        left = np.dot(self._motion_mat, covariance).transpose((1, 0, 2))
+        covariance = np.dot(left, self._motion_mat.T) + motion_cov
+
+        return mean, covariance
+
+    def update(
+        self, mean: np.ndarray, covariance: np.ndarray, measurement: np.ndarray
+    ) -> Tuple[np.ndarray, np.ndarray]:
+        """
+        Run Kalman filter correction step.
+
+        Args:
+            mean (ndarray): The predicted state's mean vector (8 dimensional).
+            covariance (ndarray): The state's covariance matrix (8x8 dimensional).
+            measurement (ndarray): The 4-dimensional measurement vector (x, y, a, h), where (x, y) is the center position, a the aspect ratio, and h the height of the bounding box.
+
+        Returns:
+            Tuple[ndarray, ndarray]: Returns the measurement-corrected state distribution.
+        """
+        projected_mean, projected_cov = self.project(mean, covariance)
+
+        chol_factor, lower = scipy.linalg.cho_factor(
+            projected_cov, lower=True, check_finite=False
+        )
+        kalman_gain = scipy.linalg.cho_solve(
+            (chol_factor, lower),
+            np.dot(covariance, self._update_mat.T).T,
+            check_finite=False,
+        ).T
+        innovation = measurement - projected_mean
+
+        new_mean = mean + np.dot(innovation, kalman_gain.T)
+        new_covariance = covariance - np.linalg.multi_dot(
+            (kalman_gain, projected_cov, kalman_gain.T)
+        )
+        return new_mean, new_covariance

supervision/tracker/byte_tracker/matching.py

@@ -0,0 +1,64 @@
+from typing import List, Tuple
+
+import numpy as np
+from scipy.optimize import linear_sum_assignment
+
+from supervision.detection.utils import box_iou_batch
+
+
+def indices_to_matches(
+    cost_matrix: np.ndarray, indices: np.ndarray, thresh: float
+) -> Tuple[np.ndarray, tuple, tuple]:
+    matched_cost = cost_matrix[tuple(zip(*indices))]
+    matched_mask = matched_cost <= thresh
+
+    matches = indices[matched_mask]
+    unmatched_a = tuple(set(range(cost_matrix.shape[0])) - set(matches[:, 0]))
+    unmatched_b = tuple(set(range(cost_matrix.shape[1])) - set(matches[:, 1]))
+    return matches, unmatched_a, unmatched_b
+
+
+def linear_assignment(
+    cost_matrix: np.ndarray, thresh: float
+) -> [np.ndarray, Tuple[int], Tuple[int, int]]:
+    if cost_matrix.size == 0:
+        return (
+            np.empty((0, 2), dtype=int),
+            tuple(range(cost_matrix.shape[0])),
+            tuple(range(cost_matrix.shape[1])),
+        )
+
+    cost_matrix[cost_matrix > thresh] = thresh + 1e-4
+    row_ind, col_ind = linear_sum_assignment(cost_matrix)
+    indices = np.column_stack((row_ind, col_ind))
+
+    return indices_to_matches(cost_matrix, indices, thresh)
+
+
+def iou_distance(atracks: List, btracks: List) -> np.ndarray:
+    if (len(atracks) > 0 and isinstance(atracks[0], np.ndarray)) or (
+        len(btracks) > 0 and isinstance(btracks[0], np.ndarray)
+    ):
+        atlbrs = atracks
+        btlbrs = btracks
+    else:
+        atlbrs = [track.tlbr for track in atracks]
+        btlbrs = [track.tlbr for track in btracks]
+
+    _ious = np.zeros((len(atlbrs), len(btlbrs)), dtype=np.float32)
+    if _ious.size != 0:
+        _ious = box_iou_batch(np.asarray(atlbrs), np.asarray(btlbrs))
+    cost_matrix = 1 - _ious
+
+    return cost_matrix
+
+
+def fuse_score(cost_matrix: np.ndarray, detections: List) -> np.ndarray:
+    if cost_matrix.size == 0:
+        return cost_matrix
+    iou_sim = 1 - cost_matrix
+    det_scores = np.array([det.score for det in detections])
+    det_scores = np.expand_dims(det_scores, axis=0).repeat(cost_matrix.shape[0], axis=0)
+    fuse_sim = iou_sim * det_scores
+    fuse_cost = 1 - fuse_sim
+    return fuse_cost

supervision/utils/file.py

@@ -0,0 +1,140 @@
+import json
+from pathlib import Path
+from typing import List, Optional, Union
+
+import numpy as np
+import yaml
+
+
+class NumpyJsonEncoder(json.JSONEncoder):
+    def default(self, obj):
+        if isinstance(obj, np.integer):
+            return int(obj)
+        if isinstance(obj, np.floating):
+            return float(obj)
+        if isinstance(obj, np.ndarray):
+            return obj.tolist()
+        return super(NumpyJsonEncoder, self).default(obj)
+
+
+def list_files_with_extensions(
+    directory: Union[str, Path], extensions: Optional[List[str]] = None
+) -> List[Path]:
+    """
+    List files in a directory with specified extensions or all files if no extensions are provided.
+
+    Args:
+        directory (Union[str, Path]): The directory path as a string or Path object.
+        extensions (Optional[List[str]]): A list of file extensions to filter. Default is None, which lists all files.
+
+    Returns:
+        (List[Path]): A list of Path objects for the matching files.
+
+    Examples:
+        ```python
+        >>> import supervision as sv
+
+        >>> # List all files in the directory
+        >>> files = sv.list_files_with_extensions(directory='my_directory')
+
+        >>> # List only files with '.txt' and '.md' extensions
+        >>> files = sv.list_files_with_extensions(directory='my_directory', extensions=['txt', 'md'])
+        ```
+    """
+    directory = Path(directory)
+    files_with_extensions = []
+
+    if extensions is not None:
+        for ext in extensions:
+            files_with_extensions.extend(directory.glob(f"*.{ext}"))
+    else:
+        files_with_extensions.extend(directory.glob("*"))
+
+    return files_with_extensions
+
+
+def read_txt_file(file_path: str) -> List[str]:
+    """
+    Read a text file and return a list of strings without newline characters.
+
+    Args:
+        file_path (str): The path to the text file.
+
+    Returns:
+        List[str]: A list of strings representing the lines in the text file.
+    """
+    with open(file_path, "r") as file:
+        lines = file.readlines()
+        lines = [line.rstrip("\n") for line in lines]
+
+    return lines
+
+
+def save_text_file(lines: List[str], file_path: str):
+    """
+    Write a list of strings to a text file, each string on a new line.
+
+    Args:
+        lines (List[str]): The list of strings to be written to the file.
+        file_path (str): The path to the text file.
+    """
+    with open(file_path, "w") as file:
+        for line in lines:
+            file.write(line + "\n")
+
+
+def read_json_file(file_path: str) -> dict:
+    """
+    Read a json file and return a dict.
+
+    Args:
+        file_path (str): The path to the json file.
+
+    Returns:
+        dict: A dict of annotations information
+    """
+    with open(file_path, "r") as file:
+        data = json.load(file)
+    return data
+
+
+def save_json_file(data: dict, file_path: str, indent: int = 3) -> None:
+    """
+    Write a dict to a json file.
+
+    Args:
+        indent:
+        data (dict): dict with unique keys and value as pair.
+        file_path (str): The path to the json file.
+    """
+    with open(file_path, "w") as fp:
+        json.dump(data, fp, cls=NumpyJsonEncoder, indent=indent)
+
+
+def read_yaml_file(file_path: str) -> dict:
+    """
+    Read a yaml file and return a dict.
+
+    Args:
+        file_path (str): The path to the yaml file.
+
+    Returns:
+        dict: A dict of content information
+    """
+    with open(file_path, "r") as file:
+        data = yaml.safe_load(file)
+    return data
+
+
+def save_yaml_file(data: dict, file_path: str) -> None:
+    """
+    Save a dict to a json file.
+
+    Args:
+        indent:
+        data (dict): dict with unique keys and value as pair.
+        file_path (str): The path to the json file.
+    """
+
+    with open(file_path, "w") as outfile:
+        yaml.dump(data, outfile, sort_keys=False, default_flow_style=None)

supervision/utils/image.py

@@ -0,0 +1,93 @@
+import os
+import shutil
+from typing import Optional
+
+import cv2
+import numpy as np
+
+
+def crop(image: np.ndarray, xyxy: np.ndarray) -> np.ndarray:
+    """
+    Crops the given image based on the given bounding box.
+
+    Args:
+        image (np.ndarray): The image to be cropped, represented as a numpy array.
+        xyxy (np.ndarray): A numpy array containing the bounding box coordinates in the format (x1, y1, x2, y2).
+
+    Returns:
+        (np.ndarray): The cropped image as a numpy array.
+
+    Examples:
+        ```python
+        >>> import supervision as sv
+
+        >>> detection = sv.Detections(...)
+        >>> with sv.ImageSink(target_dir_path='target/directory/path') as sink:
+        ...     for xyxy in detection.xyxy:
+        ...         cropped_image = sv.crop(image=image, xyxy=xyxy)
+        ...         sink.save_image(image=image)
+        ```
+    """
+
+    xyxy = np.round(xyxy).astype(int)
+    x1, y1, x2, y2 = xyxy
+    cropped_img = image[y1:y2, x1:x2]
+    return cropped_img
+
+
+class ImageSink:
+    def __init__(
+        self,
+        target_dir_path: str,
+        overwrite: bool = False,
+        image_name_pattern: str = "image_{:05d}.png",
+    ):
+        """
+        Initialize a context manager for saving images.
+
+        Args:
+            target_dir_path (str): The target directory where images will be saved.
+            overwrite (bool, optional): Whether to overwrite the existing directory. Defaults to False.
+            image_name_pattern (str, optional): The image file name pattern. Defaults to "image_{:05d}.png".
+
+        Examples:
+            ```python
+            >>> import supervision as sv
+
+            >>> with sv.ImageSink(target_dir_path='target/directory/path', overwrite=True) as sink:
+            ...     for image in sv.get_video_frames_generator(source_path='source_video.mp4', stride=2):
+            ...         sink.save_image(image=image)
+            ```
+        """
+        self.target_dir_path = target_dir_path
+        self.overwrite = overwrite
+        self.image_name_pattern = image_name_pattern
+        self.image_count = 0
+
+    def __enter__(self):
+        if os.path.exists(self.target_dir_path):
+            if self.overwrite:
+                shutil.rmtree(self.target_dir_path)
+                os.makedirs(self.target_dir_path)
+        else:
+            os.makedirs(self.target_dir_path)
+
+        return self
+
+    def save_image(self, image: np.ndarray, image_name: Optional[str] = None):
+        """
+        Save a given image in the target directory.
+
+        Args:
+            image (np.ndarray): The image to be saved.
+            image_name (str, optional): The name to use for the saved image. If not provided, a name will be generated using the `image_name_pattern`.
+        """
+        if image_name is None:
+            image_name = self.image_name_pattern.format(self.image_count)
+
+        image_path = os.path.join(self.target_dir_path, image_name)
+        cv2.imwrite(image_path, image)
+        self.image_count += 1
+
+    def __exit__(self, exc_type, exc_value, exc_traceback):
+        pass

supervision/utils/internal.py

@@ -0,0 +1,18 @@
+import functools
+import warnings
+
+
+def deprecated(reason: str):
+    def decorator(func):
+        @functools.wraps(func)
+        def wrapper(*args, **kwargs):
+            warnings.warn(
+                f"{func.__name__} is deprecated: {reason}",
+                category=DeprecationWarning,
+                stacklevel=2,
+            )
+            return func(*args, **kwargs)
+
+        return wrapper
+
+    return decorator

supervision/utils/notebook.py

@@ -24,7 +24,7 @@ def plot_image(
         >>> image = cv2.imread("path/to/image.jpg")
 
         %matplotlib inline
-        >>> sv.plot_image(image, (16, 16))
+        >>> sv.plot_image(image=image, size=(16, 16))
         ```
     """
     plt.figure(figsize=size)
@@ -71,7 +71,7 @@ def plot_images_grid(
         >>> titles = ["Image 1", "Image 2", "Image 3"]
 
         %matplotlib inline
-        >>> plot_images_grid(images, grid_size=(2, 2), titles=titles, figsize=(16, 16))
+        >>> plot_images_grid(images, grid_size=(2, 2), titles=titles, size=(16, 16))
         ```
     """
     nrows, ncols = grid_size

supervision/utils/video.py

@@ -20,9 +20,9 @@ class VideoInfo:
 
     Examples:
         ```python
-        >>> from supervision import VideoInfo
+        >>> import supervision as sv
 
-        >>> video_info = VideoInfo.from_video_path(video_path='video.mp4')
+        >>> video_info = sv.VideoInfo.from_video_path(video_path='video.mp4')
 
         >>> video_info
         VideoInfo(width=3840, height=2160, fps=25, total_frames=538)
@@ -63,15 +63,15 @@ class VideoSink:
         target_path (str): The path to the output file where the video will be saved.
         video_info (VideoInfo): Information about the video resolution, fps, and total frame count.
 
-    Examples:
+    Example:
         ```python
-        >>> from supervision import VideoInfo, VideoSink
+        >>> import supervision as sv
 
-        >>> video_info = VideoInfo.from_video_path(video_path='source_video.mp4')
+        >>> video_info = sv.VideoInfo.from_video_path(video_path='source_video.mp4')
 
-        >>> with VideoSink(target_path='target_video.mp4', video_info=video_info) as s:
-        ...     frame = ...
-        ...     s.write_frame(frame=frame)
+        >>> with sv.VideoSink(target_path='target_video.mp4', video_info=video_info) as sink:
+        ...     for frame in get_video_frames_generator(source_path='source_video.mp4', stride=2):
+        ...         sink.write_frame(frame=frame)
         ```
     """
 
@@ -93,35 +93,58 @@ class VideoSink:
     def write_frame(self, frame: np.ndarray):
         self.__writer.write(frame)
 
-    def __exit__(self, exc_type, exc_val, exc_tb):
+    def __exit__(self, exc_type, exc_value, exc_traceback):
         self.__writer.release()
 
 
-def get_video_frames_generator(source_path: str) -> Generator[np.ndarray, None, None]:
+def _validate_and_setup_video(source_path: str, start: int, end: Optional[int]):
+    video = cv2.VideoCapture(source_path)
+    if not video.isOpened():
+        raise Exception(f"Could not open video at {source_path}")
+    total_frames = int(video.get(cv2.CAP_PROP_FRAME_COUNT))
+    if end is not None and end > total_frames:
+        raise Exception(f"Requested frames are outbound")
+    start = max(start, 0)
+    end = min(end, total_frames) if end is not None else total_frames
+    video.set(cv2.CAP_PROP_POS_FRAMES, start)
+    return video, start, end
+
+
+def get_video_frames_generator(
+    source_path: str, stride: int = 1, start: int = 0, end: Optional[int] = None
+) -> Generator[np.ndarray, None, None]:
     """
     Get a generator that yields the frames of the video.
 
     Args:
         source_path (str): The path of the video file.
+        stride (int): Indicates the interval at which frames are returned, skipping stride - 1 frames between each.
+        start (int): Indicates the starting position from which video should generate frames
+        end (Optional[int]): Indicates the ending position at which video should stop generating frames. If None, video will be read to the end.
 
     Returns:
         (Generator[np.ndarray, None, None]): A generator that yields the frames of the video.
 
     Examples:
         ```python
-        >>> from supervision import get_video_frames_generator
+        >>> import supervision as sv
 
-        >>> for frame in get_video_frames_generator(source_path='source_video.mp4'):
+        >>> for frame in sv.get_video_frames_generator(source_path='source_video.mp4'):
         ...     ...
         ```
     """
-    video = cv2.VideoCapture(source_path)
-    if not video.isOpened():
-        raise Exception(f"Could not open video at {source_path}")
-    success, frame = video.read()
-    while success:
-        yield frame
+    video, start, end = _validate_and_setup_video(source_path, start, end)
+    frame_position = start
+    while True:
         success, frame = video.read()
+        if not success or frame_position >= end:
+            break
+        yield frame
+        for _ in range(stride - 1):
+            success = video.grab()
+            if not success:
+                break
+        frame_position += stride
     video.release()
 
 
@@ -140,15 +163,15 @@ def process_video(
 
     Examples:
         ```python
-        >>> from supervision import process_video
+        >>> import supervision as sv
 
-        >>> def process_frame(scene: np.ndarray) -> np.ndarray:
+        >>> def callback(scene: np.ndarray, index: int) -> np.ndarray:
         ...     ...
 
         >>> process_video(
-        ...     source_path='source_video.mp4',
-        ...     target_path='target_video.mp4',
-        ...     callback=process_frame
+        ...     source_path='...',
+        ...     target_path='...',
+        ...     callback=callback
         ... )
         ```
     """

test/classification/test_core.py

@@ -0,0 +1,63 @@
+from typing import List, Optional, Tuple
+from contextlib import ExitStack as DoesNotRaise
+
+import pytest
+import numpy as np
+
+from supervision.classification.core import Classifications
+
+
+@pytest.mark.parametrize(
+    "class_id, confidence, k, expected_result, exception",
+    [
+        (
+            np.array([0, 1, 2, 3, 4]),
+            np.array([0.1, 0.2, 0.9, 0.4, 0.5]),
+            5,
+            (np.array([2, 4, 3, 1, 0]), np.array([0.9, 0.5, 0.4, 0.2, 0.1])),
+            DoesNotRaise(),
+        ),  # class_id with 5 numbers and 5 confidences
+        (
+            np.array([5, 1, 2, 3, 4]),
+            np.array([0.1, 0.2, 0.9, 0.4, 0.5]),
+            1,
+            (np.array([2]), np.array([0.9])),
+            DoesNotRaise(),
+        ),  # class_id with 5 numbers and 5 confidences, retrieve where k = 1
+        (
+            np.array([4, 1, 2, 3, 6, 5]),
+            np.array([0.8, 0.2, 0.9, 0.4, 0.5, 0.1]),
+            2,
+            (np.array([2, 4]), np.array([0.9, 0.8])),
+            DoesNotRaise(),
+        ),  # class_id with 5 numbers and 5 confidences, retrieve where k = 3
+        (
+            np.array([0, 1, 2, 3, 4]),
+            np.array([]),
+            5,
+            None,
+            pytest.raises(ValueError),
+        ),  # class_id with 5 numbers and 0 confidences
+        (
+            [0, 1, 2, 3, 4],
+            [0.1, 0.2, 0.3, 0.4],
+            5,
+            None,
+            pytest.raises(ValueError),
+        ),  # class_id with 5 numbers and 4 confidences
+    ],
+)
+def test_top_k(
+    class_id: np.ndarray,
+    confidence: Optional[np.ndarray],
+    k: int,
+    expected_result: Optional[Tuple[np.ndarray, np.ndarray]],
+    exception: Exception,
+) -> None:
+    with exception:
+        result = Classifications(
+            class_id=np.array(class_id), confidence=np.array(confidence)
+        ).get_top_k(k)
+
+        assert np.array_equal(result[0], expected_result[0])
+        assert np.array_equal(result[1], expected_result[1])

test/dataset/formats/test_coco.py

@@ -0,0 +1,424 @@
+from contextlib import ExitStack as DoesNotRaise
+from typing import List, Tuple, Dict
+
+import pytest
+
+from supervision import Detections
+from supervision.dataset.formats.coco import classes_to_coco_categories, coco_categories_to_classes, \
+    group_coco_annotations_by_image_id, coco_annotations_to_detections, build_coco_class_index_mapping
+
+import numpy as np
+
+
+def mock_cock_coco_annotation(
+    annotation_id: int = 0,
+    image_id: int = 0,
+    category_id: int = 0,
+    bbox: Tuple[float, float, float, float] = (0.0, 0.0, 0.0, 0.0),
+    area: float = 0.0
+) -> dict:
+    return {
+        "id": annotation_id,
+        "image_id": image_id,
+        "category_id": category_id,
+        "bbox": list(bbox),
+        "area": area,
+        "iscrowd": 0
+    }
+
+
+@pytest.mark.parametrize(
+    "coco_categories, expected_result, exception",
+    [
+        (
+            [],
+            [],
+            DoesNotRaise()
+        ),  # empty coco categories
+        (
+            [
+                {
+                    "id": 0,
+                    "name": "fashion-assistant",
+                    "supercategory": "none"
+                }
+            ],
+            ["fashion-assistant"],
+            DoesNotRaise()
+        ),  # single coco category with supercategory == "none"
+        (
+            [
+                {
+                    "id": 0,
+                    "name": "fashion-assistant",
+                    "supercategory": "none"
+                },
+                {
+                    "id": 1,
+                    "name": "baseball cap",
+                    "supercategory": "fashion-assistant"
+                }
+            ],
+            [
+                "fashion-assistant", "baseball cap"
+            ],
+            DoesNotRaise()
+        ),  # two coco categories; one with supercategory == "none" and one with supercategory != "none"
+        (
+            [
+                {
+                    "id": 0,
+                    "name": "fashion-assistant",
+                    "supercategory": "none"
+                },
+                {
+                    "id": 1,
+                    "name": "baseball cap",
+                    "supercategory": "fashion-assistant"
+                },
+                {
+                    "id": 2,
+                    "name": "hoodie",
+                    "supercategory": "fashion-assistant"
+                }
+            ],
+            [
+                "fashion-assistant",
+                "baseball cap",
+                "hoodie"
+            ],
+            DoesNotRaise()
+        ),  # three coco categories; one with supercategory == "none" and two with supercategory != "none"
+        (
+            [
+                {
+                    "id": 0,
+                    "name": "fashion-assistant",
+                    "supercategory": "none"
+                },
+                {
+                    "id": 2,
+                    "name": "hoodie",
+                    "supercategory": "fashion-assistant"
+                },
+                {
+                    "id": 1,
+                    "name": "baseball cap",
+                    "supercategory": "fashion-assistant"
+                }
+            ],
+            [
+                "fashion-assistant",
+                "baseball cap",
+                "hoodie"
+            ],
+            DoesNotRaise()
+        ),  # three coco categories; one with supercategory == "none" and two with supercategory != "none" (different order)
+    ]
+)
+def test_coco_categories_to_classes(
+    coco_categories: List[dict],
+    expected_result: List[str],
+    exception: Exception
+) -> None:
+    with exception:
+        result = coco_categories_to_classes(coco_categories=coco_categories)
+        assert result == expected_result
+
+
+@pytest.mark.parametrize(
+    "classes, exception",
+    [
+        (
+            [],
+            DoesNotRaise()
+        ),  # empty classes
+        (
+            [
+                "baseball cap"
+            ],
+            DoesNotRaise()
+        ),  # single class
+        (
+            [
+                "baseball cap",
+                "hoodie"
+            ],
+            DoesNotRaise()
+        ),  # two classes
+    ]
+)
+def test_classes_to_coco_categories_and_back_to_classes(classes: List[str], exception: Exception) -> None:
+    with exception:
+        coco_categories = classes_to_coco_categories(classes=classes)
+        result = coco_categories_to_classes(coco_categories=coco_categories)
+        assert result == classes
+
+
+@pytest.mark.parametrize(
+    "coco_annotations, expected_result, exception",
+    [
+        (
+            [],
+            {},
+            DoesNotRaise()
+        ),  # empty coco annotations
+        (
+            [
+                mock_cock_coco_annotation(annotation_id=0, image_id=0, category_id=0)
+            ],
+            {
+                0: [
+                    mock_cock_coco_annotation(annotation_id=0, image_id=0, category_id=0)
+                ]
+            },
+            DoesNotRaise()
+        ),  # single coco annotation
+        (
+            [
+                mock_cock_coco_annotation(annotation_id=0, image_id=0, category_id=0),
+                mock_cock_coco_annotation(annotation_id=1, image_id=1, category_id=0)
+            ],
+            {
+                0: [
+                    mock_cock_coco_annotation(annotation_id=0, image_id=0, category_id=0)
+                ],
+                1: [
+                    mock_cock_coco_annotation(annotation_id=1, image_id=1, category_id=0)
+                ]
+            },
+            DoesNotRaise()
+        ),  # two coco annotations
+        (
+            [
+                mock_cock_coco_annotation(annotation_id=0, image_id=0, category_id=0),
+                mock_cock_coco_annotation(annotation_id=1, image_id=1, category_id=1),
+                mock_cock_coco_annotation(annotation_id=2, image_id=1, category_id=2),
+                mock_cock_coco_annotation(annotation_id=3, image_id=2, category_id=3),
+                mock_cock_coco_annotation(annotation_id=4, image_id=3, category_id=1),
+                mock_cock_coco_annotation(annotation_id=5, image_id=3, category_id=2),
+                mock_cock_coco_annotation(annotation_id=5, image_id=3, category_id=3),
+            ],
+            {
+                0: [
+                    mock_cock_coco_annotation(annotation_id=0, image_id=0, category_id=0),
+                ],
+                1: [
+                    mock_cock_coco_annotation(annotation_id=1, image_id=1, category_id=1),
+                    mock_cock_coco_annotation(annotation_id=2, image_id=1, category_id=2),
+                ],
+                2: [
+                    mock_cock_coco_annotation(annotation_id=3, image_id=2, category_id=3),
+                ],
+                3: [
+                    mock_cock_coco_annotation(annotation_id=4, image_id=3, category_id=1),
+                    mock_cock_coco_annotation(annotation_id=5, image_id=3, category_id=2),
+                    mock_cock_coco_annotation(annotation_id=5, image_id=3, category_id=3),
+                ]
+            },
+            DoesNotRaise()
+        ),  # two coco annotations
+    ]
+)
+def test_group_coco_annotations_by_image_id(
+    coco_annotations: List[dict],
+    expected_result: dict,
+    exception: Exception
+) -> None:
+    with exception:
+        result = group_coco_annotations_by_image_id(coco_annotations=coco_annotations)
+        assert result == expected_result
+
+
+@pytest.mark.parametrize(
+    "image_annotations, resolution_wh, with_masks, expected_result, exception",
+    [
+        (
+            [],
+            (1000, 1000),
+            False,
+            Detections.empty(),
+            DoesNotRaise()
+        ),  # empty image annotations
+        (
+            [
+                mock_cock_coco_annotation(category_id=0, bbox=(0, 0, 100, 100), area=100 * 100)
+            ],
+            (1000, 1000),
+            False,
+            Detections(
+                xyxy=np.array([
+                    [          0,           0,         100,         100]
+                ], dtype=np.float32),
+                class_id=np.array([
+                    0
+                ], dtype=int)
+            ),
+            DoesNotRaise()
+        ),  # single image annotations
+        (
+            [
+                mock_cock_coco_annotation(category_id=0, bbox=(0, 0, 100, 100), area=100 * 100),
+                mock_cock_coco_annotation(category_id=0, bbox=(100, 100, 100, 100), area=100 * 100),
+            ],
+            (1000, 1000),
+            False,
+            Detections(
+                xyxy=np.array([
+                    [          0,           0,         100,         100],
+                    [        100,         100,         200,         200]
+                ], dtype=np.float32),
+                class_id=np.array([
+                    0, 0
+                ], dtype=int)
+            ),
+            DoesNotRaise()
+        ),  # two image annotations
+    ]
+)
+def test_coco_annotations_to_detections(
+    image_annotations: List[dict],
+    resolution_wh: Tuple[int, int],
+    with_masks: bool,
+    expected_result: Detections,
+    exception: Exception
+) -> None:
+    with exception:
+        result = coco_annotations_to_detections(
+            image_annotations=image_annotations,
+            resolution_wh=resolution_wh,
+            with_masks=with_masks
+        )
+        assert result == expected_result
+
+@pytest.mark.parametrize(
+    "coco_categories, target_classes, expected_result, exception",
+    [
+        (
+            [],
+            [],
+            {},
+            DoesNotRaise()
+        ), # empty coco categories
+        (
+            [
+                {
+                    "id": 0,
+                    "name": "fashion-assistant",
+                    "supercategory": "none"
+                }
+            ],
+            [
+                "fashion-assistant"
+            ],
+            {
+                0: 0
+            },
+            DoesNotRaise()
+        ),  # single coco category starting from 0
+        (
+            [
+                {
+                    "id": 1,
+                    "name": "fashion-assistant",
+                    "supercategory": "none"
+                }
+            ],
+            [
+                "fashion-assistant"
+            ],
+            {
+                1: 0
+            },
+            DoesNotRaise()
+        ),  # single coco category starting from 1
+        (
+            [
+                {
+                    "id": 0,
+                    "name": "fashion-assistant",
+                    "supercategory": "none"
+                },
+                {
+                    "id": 2,
+                    "name": "hoodie",
+                    "supercategory": "fashion-assistant"
+                },
+                {
+                    "id": 1,
+                    "name": "baseball cap",
+                    "supercategory": "fashion-assistant"
+                }
+            ],
+            [
+                "fashion-assistant",
+                "baseball cap",
+                "hoodie"
+            ],
+            {
+                0: 0,
+                1: 1,
+                2: 2
+            },
+            DoesNotRaise()
+        ),  # three coco categories
+        (
+            [
+                {
+                    "id": 2,
+                    "name": "hoodie",
+                    "supercategory": "fashion-assistant"
+                },
+                {
+                    "id": 1,
+                    "name": "baseball cap",
+                    "supercategory": "fashion-assistant"
+                }
+            ],
+            [
+                "baseball cap",
+                "hoodie"
+            ],
+            {
+                2: 1,
+                1: 0
+            },
+            DoesNotRaise()
+        ),  # two coco categories
+        (
+            [
+                {
+                    "id": 3,
+                    "name": "hoodie",
+                    "supercategory": "fashion-assistant"
+                },
+                {
+                    "id": 1,
+                    "name": "baseball cap",
+                    "supercategory": "fashion-assistant"
+                }
+            ],
+            [
+                "baseball cap",
+                "hoodie"
+            ],
+            {
+                3: 1,
+                1: 0
+            },
+            DoesNotRaise()
+        ),  # two coco categories with missing category
+    ]
+)
+def test_build_coco_class_index_mapping(
+    coco_categories: List[dict],
+    target_classes: List[str],
+    expected_result: Dict[int, int],
+    exception: Exception
+) -> None:
+    with exception:
+        result = build_coco_class_index_mapping(
+            coco_categories=coco_categories,
+            target_classes=target_classes
+        )
+        assert result == expected_result

test/dataset/formats/test_pascal_voc.py

@@ -0,0 +1,147 @@
+import xml.etree.ElementTree as ET
+from contextlib import ExitStack as DoesNotRaise
+from test.utils import mock_detections
+from typing import List, Optional
+
+import numpy as np
+import pytest
+
+from supervision.dataset.formats.pascal_voc import (
+    detections_from_xml_obj,
+    object_to_pascal_voc,
+    parse_polygon_points,
+)
+
+
+def are_xml_elements_equal(elem1, elem2):
+    if (
+        elem1.tag != elem2.tag
+        or elem1.attrib != elem2.attrib
+        or elem1.text != elem2.text
+        or len(elem1) != len(elem2)
+    ):
+        return False
+
+    for child1, child2 in zip(elem1, elem2):
+        if not are_xml_elements_equal(child1, child2):
+            return False
+
+    return True
+
+
+@pytest.mark.parametrize(
+    "xyxy, name, polygon, expected_result, exception",
+    [
+        (
+            [0, 0, 10, 10],
+            "test",
+            None,
+            ET.fromstring(
+                """<object><name>test</name><bndbox><xmin>0</xmin><ymin>0</ymin><xmax>10</xmax><ymax>10</ymax></bndbox></object>"""
+            ),
+            DoesNotRaise(),
+        ),
+        (
+            [0, 0, 10, 10],
+            "test",
+            [[0, 0], [10, 0], [10, 10], [0, 10]],
+            ET.fromstring(
+                """<object><name>test</name><bndbox><xmin>0</xmin><ymin>0</ymin><xmax>10</xmax><ymax>10</ymax></bndbox><polygon><x1>0</x1><y1>0</y1><x2>10</x2><y2>0</y2><x3>10</x3><y3>10</y3><x4>0</x4><y4>10</y4></polygon></object>"""
+            ),
+            DoesNotRaise(),
+        ),
+    ],
+)
+def test_object_to_pascal_voc(
+    xyxy: np.ndarray,
+    name: str,
+    polygon: Optional[np.ndarray],
+    expected_result,
+    exception: Exception,
+):
+    with exception:
+        result = object_to_pascal_voc(xyxy=xyxy, name=name, polygon=polygon)
+        assert are_xml_elements_equal(result, expected_result)
+
+
+@pytest.mark.parametrize(
+    "polygon_element, expected_result, exception",
+    [
+        (
+            ET.fromstring(
+                """<polygon><x1>0</x1><y1>0</y1><x2>10</x2><y2>0</y2><x3>10</x3><y3>10</y3><x4>0</x4><y4>10</y4></polygon>"""
+            ),
+            [[0, 0], [10, 0], [10, 10], [0, 10]],
+            DoesNotRaise(),
+        )
+    ],
+)
+def test_parse_polygon_points(
+    polygon_element,
+    expected_result: List[list],
+    exception,
+):
+    with exception:
+        result = parse_polygon_points(polygon_element)
+        assert result == expected_result
+
+
+ONE_CLASS_N_BBOX = """<annotation><object><name>test</name><bndbox><xmin>0</xmin><ymin>0</ymin><xmax>10</xmax><ymax>10</ymax></bndbox></object><object><name>test</name><bndbox><xmin>10</xmin><ymin>10</ymin><xmax>20</xmax><ymax>20</ymax></bndbox></object></annotation>"""
+
+
+ONE_CLASS_ONE_BBOX = """<annotation><object><name>test</name><bndbox><xmin>0</xmin><ymin>0</ymin><xmax>10</xmax><ymax>10</ymax></bndbox></object></annotation>"""
+
+
+N_CLASS_N_BBOX = """<annotation><object><name>test</name><bndbox><xmin>0</xmin><ymin>0</ymin><xmax>10</xmax><ymax>10</ymax></bndbox></object><object><name>test</name><bndbox><xmin>20</xmin><ymin>30</ymin><xmax>30</xmax><ymax>40</ymax></bndbox></object><object><name>test2</name><bndbox><xmin>10</xmin><ymin>10</ymin><xmax>20</xmax><ymax>20</ymax></bndbox></object></annotation>"""
+
+NO_DETECTIONS = """<annotation></annotation>"""
+
+
+@pytest.mark.parametrize(
+    "xml_string, classes, resolution_wh, force_masks, expected_result, exception",
+    [
+        (
+            ONE_CLASS_ONE_BBOX,
+            ["test"],
+            (100, 100),
+            False,
+            mock_detections(np.array([[0, 0, 10, 10]]), None, [0]),
+            DoesNotRaise(),
+        ),
+        (
+            ONE_CLASS_N_BBOX,
+            ["test"],
+            (100, 100),
+            False,
+            mock_detections(np.array([[0, 0, 10, 10], [10, 10, 20, 20]]), None, [0, 0]),
+            DoesNotRaise(),
+        ),
+        (
+            N_CLASS_N_BBOX,
+            ["test", "test2"],
+            (100, 100),
+            False,
+            mock_detections(
+                np.array([[0, 0, 10, 10], [20, 30, 30, 40], [10, 10, 20, 20]]),
+                None,
+                [0, 0, 1],
+            ),
+            DoesNotRaise(),
+        ),
+        (
+            NO_DETECTIONS,
+            [],
+            (100, 100),
+            False,
+            mock_detections(np.empty((0, 4)), None, []),
+            DoesNotRaise(),
+        ),
+    ],
+)
+def test_detections_from_xml_obj(
+    xml_string, classes, resolution_wh, force_masks, expected_result, exception
+):
+    with exception:
+        root = ET.fromstring(xml_string)
+        result, _ = detections_from_xml_obj(root, classes, resolution_wh, force_masks)
+        assert result == expected_result

test/dataset/formats/test_yolo.py

@@ -0,0 +1,304 @@
+from contextlib import ExitStack as DoesNotRaise
+from typing import List, Tuple, Optional
+
+import pytest
+import numpy as np
+
+from supervision.detection.core import Detections
+from supervision.dataset.formats.yolo import yolo_annotations_to_detections, _with_mask, _image_name_to_annotation_name, \
+    object_to_yolo
+
+
+def _mock_simple_mask(resolution_wh: Tuple[int, int], box: List[int]) -> np.array:
+    x_min, y_min, x_max, y_max = box
+    mask = np.full(resolution_wh, False, dtype=bool)
+    mask[y_min:y_max, x_min:x_max] = True
+    return mask
+
+
+# The result of _mock_simple_mask is a little different from the result produced by cv2.
+def _arrays_almost_equal(arr1: np.ndarray, arr2: np.ndarray, threshold: float = 0.99) -> bool:
+    equal_elements = np.equal(arr1, arr2)
+    proportion_equal = np.mean(equal_elements)
+    return proportion_equal >= threshold
+
+
+@pytest.mark.parametrize(
+    'lines, expected_result, exception',
+    [
+        (
+            [],
+            False,
+            DoesNotRaise()
+        ),  # empty yolo annotation file
+        (
+            [
+                '0 0.5 0.5 0.2 0.2'
+            ],
+            False,
+            DoesNotRaise()
+        ),  # yolo annotation file with single line with box
+        (
+                [
+                    '0 0.50 0.50 0.20 0.20',
+                    '1 0.11 0.47 0.22 0.30'
+                ],
+                False,
+                DoesNotRaise()
+        ),  # yolo annotation file with two lines with box
+        (
+            [
+                '0 0.5 0.5 0.2 0.2'
+            ],
+            False,
+            DoesNotRaise()
+        ),
+        (
+            [
+                '0 0.4 0.4 0.6 0.4 0.6 0.6 0.4 0.6'
+            ],
+            True,
+            DoesNotRaise()
+        ),  # yolo annotation file with single line with polygon
+        (
+            [
+                '0 0.4 0.4 0.6 0.4 0.6 0.6 0.4 0.6',
+                '1 0.11 0.47 0.22 0.30'
+            ],
+            True,
+            DoesNotRaise()
+        ),  # yolo annotation file with two lines - one box and one polygon
+    ]
+)
+def test_with_mask(lines: List[str], expected_result: Optional[bool], exception: Exception) -> None:
+    with exception:
+        result = _with_mask(lines=lines)
+        assert result == expected_result
+
+
+@pytest.mark.parametrize(
+    'lines, resolution_wh, with_masks, expected_result, exception',
+    [
+        (
+            [],
+            (1000, 1000),
+            False,
+            Detections.empty(),
+            DoesNotRaise()
+        ),  # empty yolo annotation file
+        (
+            [
+                '0 0.5 0.5 0.2 0.2'
+            ],
+            (1000, 1000),
+            False,
+            Detections(
+                xyxy=np.array([
+                    [400, 400, 600, 600]
+                ], dtype=np.float32),
+                class_id=np.array([0], dtype=int)
+            ),
+            DoesNotRaise()
+        ),  # yolo annotation file with single line with box
+        (
+            [
+                '0 0.50 0.50 0.20 0.20',
+                '1 0.11 0.47 0.22 0.30'
+            ],
+            (1000, 1000),
+            False,
+            Detections(
+                xyxy=np.array([
+                    [400, 400, 600, 600],
+                    [  0, 320, 220, 620]
+                ], dtype=np.float32),
+                class_id=np.array([0, 1], dtype=int)
+            ),
+            DoesNotRaise()
+        ),  # yolo annotation file with two lines with box
+        (
+            [
+                '0 0.5 0.5 0.2 0.2'
+            ],
+            (1000, 1000),
+            True,
+            Detections(
+                xyxy=np.array([
+                    [400, 400, 600, 600]
+                ], dtype=np.float32),
+                class_id=np.array([0], dtype=int),
+                mask=np.array([
+                    _mock_simple_mask(resolution_wh=(1000, 1000), box=[400, 400, 600, 600])
+                ], dtype=bool)
+            ),
+            DoesNotRaise()
+        ),  # yolo annotation file with single line with box in with_masks mode
+        (
+            [
+                '0 0.4 0.4 0.6 0.4 0.6 0.6 0.4 0.6'
+            ],
+            (1000, 1000),
+            True,
+            Detections(
+                xyxy=np.array([
+                    [400, 400, 600, 600]
+                ], dtype=np.float32),
+                class_id=np.array([0], dtype=int),
+                mask=np.array([
+                    _mock_simple_mask(resolution_wh=(1000, 1000), box=[400, 400, 600, 600])
+                ], dtype=bool)
+            ),
+            DoesNotRaise()
+        ),  # yolo annotation file with single line with polygon
+        (
+            [
+                '0 0.4 0.4 0.6 0.4 0.6 0.6 0.4 0.6',
+                '1 0.11 0.47 0.22 0.30'
+            ],
+            (1000, 1000),
+            True,
+            Detections(
+                xyxy=np.array([
+                    [400, 400, 600, 600],
+                    [  0, 320, 220, 620]
+                ], dtype=np.float32),
+                class_id=np.array([0, 1], dtype=int),
+                mask=np.array([
+                    _mock_simple_mask(resolution_wh=(1000, 1000), box=[400, 400, 600, 600]),
+                    _mock_simple_mask(resolution_wh=(1000, 1000), box=[  0, 320, 220, 620])
+                ], dtype=bool)
+            ),
+            DoesNotRaise()
+        ),  # yolo annotation file with two lines - one box and one polygon in with_masks mode
+        (
+            [
+                '0 0.4 0.4 0.6 0.4 0.6 0.6 0.4 0.6',
+                '1 0.11 0.47 0.22 0.30'
+            ],
+            (1000, 1000),
+            False,
+            Detections(
+                xyxy=np.array([
+                    [400, 400, 600, 600],
+                    [  0, 320, 220, 620]
+                ], dtype=np.float32),
+                class_id=np.array([0, 1], dtype=int)
+            ),
+            DoesNotRaise()
+        ),  # yolo annotation file with two lines - one box and one polygon
+    ]
+)
+def test_yolo_annotations_to_detections(
+    lines: List[str],
+    resolution_wh: Tuple[int, int],
+    with_masks: bool,
+    expected_result: Optional[Detections],
+    exception: Exception
+) -> None:
+    with exception:
+        result = yolo_annotations_to_detections(
+            lines=lines,
+            resolution_wh=resolution_wh,
+            with_masks=with_masks)
+        assert np.array_equal(result.xyxy, expected_result.xyxy)
+        assert np.array_equal(result.class_id, expected_result.class_id)
+        assert (result.mask is None and expected_result.mask is None) or _arrays_almost_equal(result.mask, expected_result.mask)
+
+
+@pytest.mark.parametrize(
+    'image_name, expected_result, exception',
+    [
+        (
+            'image.png',
+            'image.txt',
+            DoesNotRaise()
+        ),  # simple png image
+        (
+            'image.jpeg',
+            'image.txt',
+            DoesNotRaise()
+        ),  # simple jpeg image
+        (
+            'image.jpg',
+            'image.txt',
+            DoesNotRaise()
+        ),  # simple jpg image
+        (
+            'image.000.jpg',
+            'image.000.txt',
+            DoesNotRaise()
+        ),  # jpg image with multiple dots in name
+    ]
+)
+def test_image_name_to_annotation_name(
+    image_name: str,
+    expected_result: Optional[str],
+    exception: Exception
+) -> None:
+    with exception:
+        result = _image_name_to_annotation_name(image_name=image_name)
+        assert result == expected_result
+
+
+@pytest.mark.parametrize(
+    'xyxy, class_id, image_shape, polygon, expected_result, exception',
+    [
+        (
+            np.array([100, 100, 200, 200], dtype=np.float32),
+            1,
+            (1000, 1000, 3),
+            None,
+            '1 0.15000 0.15000 0.10000 0.10000',
+            DoesNotRaise()
+        ),  # square bounding box on square image
+        (
+            np.array([100, 100, 200, 200], dtype=np.float32),
+            1,
+            (800, 1000, 3),
+            None,
+            '1 0.15000 0.18750 0.10000 0.12500',
+            DoesNotRaise()
+        ),  # square bounding box on horizontal image
+        (
+            np.array([100, 100, 200, 200], dtype=np.float32),
+            1,
+            (1000, 800, 3),
+            None,
+            '1 0.18750 0.15000 0.12500 0.10000',
+            DoesNotRaise()
+        ),  # square bounding box on vertical image
+        (
+            np.array([100, 200, 200, 400], dtype=np.float32),
+            1,
+            (1000, 1000, 3),
+            None,
+            '1 0.15000 0.30000 0.10000 0.20000',
+            DoesNotRaise()
+        ),  # horizontal bounding box on square image
+        (
+            np.array([200, 100, 400, 200], dtype=np.float32),
+            1,
+            (1000, 1000, 3),
+            None,
+            '1 0.30000 0.15000 0.20000 0.10000',
+            DoesNotRaise()
+        ),  # vertical bounding box on square image
+        (
+            np.array([100, 100, 200, 200], dtype=np.float32),
+            1,
+            (1000, 1000, 3),
+            np.array([
+                [100, 100],
+                [200, 100],
+                [200, 200],
+                [100, 100]
+            ], dtype=np.float32),
+            '1 0.10000 0.10000 0.20000 0.10000 0.20000 0.20000 0.10000 0.10000',
+            DoesNotRaise()
+        ),  # square mask on square image
+    ]
+)
+def test_object_to_yolo(xyxy: np.ndarray, class_id: int, image_shape: Tuple[int, int, int], polygon: Optional[np.ndarray], expected_result: Optional[str], exception: Exception) -> None:
+    with exception:
+        result = object_to_yolo(xyxy=xyxy, class_id=class_id, image_shape=image_shape, polygon=polygon)
+        assert result == expected_result

test/dataset/test_core.py

@@ -0,0 +1,175 @@
+from typing import List, Optional
+
+import pytest
+
+from supervision import DetectionDataset
+from contextlib import ExitStack as DoesNotRaise
+
+import numpy as np
+
+from test.utils import mock_detections
+
+
+@pytest.mark.parametrize(
+    "dataset_list, expected_result, exception",
+    [
+        (
+            [],
+            DetectionDataset(classes=[], images={}, annotations={}),
+            DoesNotRaise()
+        ),  # empty dataset list
+        (
+            [
+                DetectionDataset(classes=[], images={}, annotations={})
+            ],
+            DetectionDataset(classes=[], images={}, annotations={}),
+            DoesNotRaise()
+        ),  # single empty dataset
+        (
+            [
+                DetectionDataset(classes=['dog', 'person'], images={}, annotations={}),
+                DetectionDataset(classes=['dog', 'person'], images={}, annotations={})
+            ],
+            DetectionDataset(classes=['dog', 'person'], images={}, annotations={}),
+            DoesNotRaise()
+        ),  # two datasets; no images and annotations, the same classes
+        (
+            [
+                DetectionDataset(classes=['dog', 'person'], images={}, annotations={}),
+                DetectionDataset(classes=['cat'], images={}, annotations={})
+            ],
+            DetectionDataset(classes=['cat', 'dog', 'person'], images={}, annotations={}),
+            DoesNotRaise()
+        ),  # two datasets; no images and annotations, different classes
+        (
+            [
+                DetectionDataset(
+                    classes=['dog', 'person'],
+                    images={
+                        'image-1.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                        'image-2.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    },
+                    annotations={
+                        'image-1.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+                        'image-2.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+                    }
+                ),
+                DetectionDataset(classes=[], images={}, annotations={}),
+            ],
+            DetectionDataset(
+                classes=['dog', 'person'],
+                images={
+                    'image-1.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    'image-2.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                },
+                annotations={
+                    'image-1.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+                    'image-2.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+                }
+            ),
+            DoesNotRaise()
+        ),  # two datasets; images and annotations, the same classes
+        (
+            [
+                DetectionDataset(
+                    classes=['dog', 'person'],
+                    images={
+                        'image-1.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                        'image-2.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    },
+                    annotations={
+                        'image-1.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+                        'image-2.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+                    }
+                ),
+                DetectionDataset(classes=['cat'], images={}, annotations={}),
+            ],
+            DetectionDataset(
+                classes=['cat', 'dog', 'person'],
+                images={
+                    'image-1.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    'image-2.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                },
+                annotations={
+                    'image-1.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+                    'image-2.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[2]),
+                }
+            ),
+            DoesNotRaise()
+        ),  # two datasets; images and annotations, different classes
+        (
+            [
+                DetectionDataset(
+                    classes=['dog', 'person'],
+                    images={
+                        'image-1.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                        'image-2.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    },
+                    annotations={
+                        'image-1.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+                        'image-2.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+                    }
+                ),
+                DetectionDataset(
+                    classes=['cat'],
+                    images={
+                        'image-3.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    },
+                    annotations={
+                        'image-3.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+                    }
+                ),
+            ],
+            DetectionDataset(
+                classes=['cat', 'dog', 'person'],
+                images={
+                    'image-1.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    'image-2.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    'image-3.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                },
+                annotations={
+                    'image-1.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+                    'image-2.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[2]),
+                    'image-3.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+                }
+            ),
+            DoesNotRaise()
+        ),  # two datasets; images and annotations, different classes
+        (
+            [
+                DetectionDataset(
+                    classes=['dog', 'person'],
+                    images={
+                        'image-1.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                        'image-2.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    },
+                    annotations={
+                        'image-1.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+                        'image-2.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+                    }
+                ),
+                DetectionDataset(
+                    classes=['dog', 'person'],
+                    images={
+                        'image-2.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                        'image-3.png': np.zeros((100, 100, 3), dtype=np.uint8),
+                    },
+                    annotations={
+                        'image-2.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+                        'image-3.png': mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+                    }
+                ),
+            ],
+            None,
+            pytest.raises(ValueError)
+        )
+    ]
+)
+def test_dataset_merge(
+    dataset_list: List[DetectionDataset],
+    expected_result: Optional[DetectionDataset],
+    exception: Exception
+) -> None:
+    with exception:
+        result = DetectionDataset.merge(dataset_list=dataset_list)
+        assert result == expected_result

test/dataset/test_utils.py

@@ -0,0 +1,247 @@
+from contextlib import ExitStack as DoesNotRaise
+from typing import List, TypeVar, Optional, Tuple, Dict
+
+import pytest
+
+from supervision import Detections
+from supervision.dataset.utils import train_test_split, merge_class_lists, build_class_index_mapping, \
+    map_detections_class_id
+from test.utils import mock_detections
+
+T = TypeVar("T")
+
+
+@pytest.mark.parametrize(
+    'data, train_ratio, random_state, shuffle, expected_result, exception',
+    [
+        (
+            [],
+            0.5,
+            None,
+            False,
+            ([], []),
+            DoesNotRaise()
+        ),  # empty data
+        (
+            [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
+            0.5,
+            None,
+            False,
+            ([0, 1, 2, 3, 4], [5, 6, 7, 8, 9]),
+            DoesNotRaise()
+        ),  # data with 10 numbers and 50% train split
+        (
+            [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
+            1.0,
+            None,
+            False,
+            ([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], []),
+            DoesNotRaise()
+        ),  # data with 10 numbers and 100% train split
+        (
+            [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
+            0.0,
+            None,
+            False,
+            ([], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]),
+            DoesNotRaise()
+        ),  # data with 10 numbers and 0% train split
+        (
+            ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j'],
+            0.5,
+            None,
+            False,
+            (['a', 'b', 'c', 'd', 'e'], ['f', 'g', 'h', 'i', 'j']),
+            DoesNotRaise()
+        ),  # data with 10 chars and 50% train split
+        (
+            [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
+            0.5,
+            23,
+            True,
+            ([7, 8, 5, 6, 3], [2, 9, 0, 1, 4]),
+            DoesNotRaise()
+        ),  # data with 10 numbers and 50% train split with 23 random seed
+        (
+            [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
+            0.5,
+            32,
+            True,
+            ([4, 6, 0, 8, 9], [5, 7, 2, 3, 1]),
+            DoesNotRaise()
+        ),  # data with 10 numbers and 50% train split with 23 random seed
+    ]
+)
+def test_train_test_split(
+    data: List[T],
+    train_ratio: float,
+    random_state: int,
+    shuffle: bool,
+    expected_result: Optional[Tuple[List[T], List[T]]],
+    exception: Exception
+) -> None:
+    with exception:
+        result = train_test_split(data=data, train_ratio=train_ratio, random_state=random_state, shuffle=shuffle)
+        assert result == expected_result
+
+
+@pytest.mark.parametrize(
+    'class_lists, expected_result, exception',
+    [
+        (
+            [],
+            [],
+            DoesNotRaise()
+        ),  # empty class lists
+        (
+            [
+                ['dog', 'person']
+            ],
+            ['dog', 'person'],
+            DoesNotRaise()
+        ),  # single class list; already alphabetically sorted
+        (
+            [
+                ['person', 'dog']
+            ],
+            ['dog', 'person'],
+            DoesNotRaise()
+        ),  # single class list; not alphabetically sorted
+        (
+            [
+                ['dog', 'person'],
+                ['dog', 'person']
+            ],
+            ['dog', 'person'],
+            DoesNotRaise()
+        ),  # two class lists; the same classes; already alphabetically sorted
+        (
+            [
+                ['dog', 'person'],
+                ['cat']
+            ],
+            ['cat', 'dog', 'person'],
+            DoesNotRaise()
+        ),  # two class lists; different classes; already alphabetically sorted
+    ]
+)
+def test_merge_class_maps(class_lists: List[List[str]], expected_result: List[str], exception: Exception) -> None:
+    with exception:
+        result = merge_class_lists(class_lists=class_lists)
+        assert result == expected_result
+
+
+@pytest.mark.parametrize(
+    'source_classes, target_classes, expected_result, exception',
+    [
+        (
+            [],
+            [],
+            {},
+            DoesNotRaise()
+        ),  # empty class lists
+        (
+            [],
+            ['dog', 'person'],
+            {},
+            DoesNotRaise()
+        ),  # empty source class list
+        (
+            ['dog', 'person'],
+            [],
+            None,
+            pytest.raises(ValueError)
+        ),  # empty target class list
+        (
+            ['dog', 'person'],
+            ['dog', 'person'],
+            {0: 0, 1: 1},
+            DoesNotRaise()
+        ),  # same class lists
+        (
+            ['dog', 'person'],
+            ['person', 'dog'],
+            {0: 1, 1: 0},
+            DoesNotRaise()
+        ),  # same class lists but not alphabetically sorted
+        (
+            ['dog', 'person'],
+            ['cat', 'dog', 'person'],
+            {0: 1, 1: 2},
+            DoesNotRaise()
+        ),  # source class list is a subset of target class list
+        (
+            ['dog', 'person'],
+            ['cat', 'dog'],
+            None,
+            pytest.raises(ValueError)
+        ),  # source class list is not a subset of target class list
+    ]
+)
+def test_build_class_index_mapping(
+    source_classes: List[str],
+    target_classes: List[str],
+    expected_result: Optional[Dict[int, int]],
+    exception: Exception
+) -> None:
+    with exception:
+        result = build_class_index_mapping(source_classes=source_classes, target_classes=target_classes)
+        assert result == expected_result
+
+
+@pytest.mark.parametrize(
+    'source_to_target_mapping, detections, expected_result, exception',
+    [
+        (
+            {},
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+            None,
+            pytest.raises(ValueError)
+        ),  # empty mapping
+        (
+            {0: 1},
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+            DoesNotRaise()
+        ),  # single mapping
+        (
+                {0: 1, 1: 2},
+                Detections.empty(),
+                Detections.empty(),
+                DoesNotRaise()
+        ),  # empty detections
+        (
+            {0: 1, 1: 2},
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0]),
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1]),
+            DoesNotRaise()
+        ),  # multiple mappings
+        (
+            {0: 1, 1: 2},
+            mock_detections(xyxy=[[0, 0, 10, 10], [0, 0, 10, 10]], class_id=[0, 1]),
+            mock_detections(xyxy=[[0, 0, 10, 10], [0, 0, 10, 10]], class_id=[1, 2]),
+            DoesNotRaise()
+        ),  # multiple mappings
+        (
+            {0: 1, 1: 2},
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[2]),
+            None,
+            pytest.raises(ValueError)
+        ),  # class_id not in mapping
+        (
+            {0: 1, 1: 2},
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0], confidence=[0.5]),
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[1], confidence=[0.5]),
+            DoesNotRaise(),
+        ),  # confidence is not None
+    ]
+)
+def test_map_detections_class_id(
+    source_to_target_mapping: Dict[int, int],
+    detections: Detections,
+    expected_result: Optional[Detections],
+    exception: Exception
+) -> None:
+    with exception:
+        result = map_detections_class_id(source_to_target_mapping=source_to_target_mapping, detections=detections)
+        assert result == expected_result

test/detection/test_core.py

@@ -4,10 +4,11 @@ import pytest
 
 from supervision import Detections
 
-from typing import Optional, Union
+from typing import Optional, Union, List
 
 import numpy as np
 
+from test.utils import mock_detections
 
 PREDICTIONS = np.array([
         [       2254,         906,        2447,        1353,     0.90538,           0],
@@ -28,51 +29,30 @@ DETECTIONS = Detections(
 )
 
 
-PREDICTIONS = np.array([
-        [       2254,         906,        2447,        1353,     0.90538,           0],
-        [       2049,        1133,        2226,        1371,     0.59002,          56],
-        [        727,        1224,         838,        1601,     0.51119,          39],
-    ], dtype=np.float32)
-
-
 @pytest.mark.parametrize(
     'detections, index, expected_result, exception',
     [
         (
             DETECTIONS,
             DETECTIONS.class_id == 0,
-            Detections(
-                xyxy=np.array([
-                    [       2254,         906,        2447,        1353]
-                ], dtype=np.float32),
-                confidence=np.array([
-                    0.90538
-                ], dtype=np.float32),
-                class_id=np.array([
-                    0
-                ], dtype=int)
+            mock_detections(
+                xyxy=[[2254, 906, 2447, 1353]],
+                confidence=[0.90538],
+                class_id=[0]
             ),
             DoesNotRaise()
         ),  # take only detections with class_id = 0
         (
             DETECTIONS,
             DETECTIONS.confidence > 0.5,
-            Detections(
-                xyxy=np.array([
-                    [       2254,         906,        2447,        1353],
-                    [       2049,        1133,        2226,        1371],
-                    [        727,        1224,         838,        1601]
-                ], dtype=np.float32),
-                confidence=np.array([
-                    0.90538,
-                    0.59002,
-                    0.51119
-                ], dtype=np.float32),
-                class_id=np.array([
-                    0,
-                    56,
-                    39
-                ], dtype=int)
+            mock_detections(
+                xyxy=[
+                    [2254, 906, 2447, 1353],
+                    [2049, 1133, 2226, 1371],
+                    [727, 1224, 838, 1601]
+                ],
+                confidence=[0.90538, 0.59002, 0.51119],
+                class_id=[0, 56, 39]
             ),
             DoesNotRaise()
         ),  # take only detections with confidence > 0.5
@@ -92,14 +72,174 @@ PREDICTIONS = np.array([
             ),
             DoesNotRaise()
         ),  # take no detections
+        (
+            DETECTIONS,
+            [0, 2],
+            mock_detections(
+                xyxy=[
+                    [2254, 906, 2447, 1353],
+                    [727, 1224, 838, 1601]
+                ],
+                confidence=[0.90538, 0.51119],
+                class_id=[0, 39]
+            ),
+            DoesNotRaise()
+        ),  # take only first and third detection using List[int] index
+        (
+            DETECTIONS,
+            np.array([0, 2]),
+            mock_detections(
+                xyxy=[
+                    [2254, 906, 2447, 1353],
+                    [727, 1224, 838, 1601]
+                ],
+                confidence=[0.90538, 0.51119],
+                class_id=[0, 39]
+            ),
+            DoesNotRaise()
+        ),  # take only first and third detection using np.ndarray index
+        (
+            DETECTIONS,
+            0,
+            mock_detections(
+                xyxy=[[2254, 906, 2447, 1353]],
+                confidence=[0.90538],
+                class_id=[0]
+            ),
+            DoesNotRaise()
+        ),  # take only first detection by index
+        (
+            DETECTIONS,
+            slice(1, 3),
+            mock_detections(
+                xyxy=[
+                    [2049, 1133, 2226, 1371],
+                    [727, 1224, 838, 1601]
+                ],
+                confidence=[0.59002, 0.51119],
+                class_id=[56, 39]
+            ),
+            DoesNotRaise()
+        ),  # take only first detection by index slice (1, 3)
+        (
+            DETECTIONS,
+            10,
+            None,
+            pytest.raises(IndexError)
+        ),  # index out of range
+        (
+            DETECTIONS,
+            [0, 2, 10],
+            None,
+            pytest.raises(IndexError)
+        ),  # index out of range
+        (
+            DETECTIONS,
+            np.array([0, 2, 10]),
+            None,
+            pytest.raises(IndexError)
+        ),
+        (
+            DETECTIONS,
+            np.array([True, True, True, True, True, True, True, True, True, True, True]),
+            None,
+            pytest.raises(IndexError)
+        )
     ]
 )
 def test_getitem(
         detections: Detections,
-        index: Union[int, slice, np.ndarray],
+        index:  Union[int, slice, List[int], np.ndarray],
         expected_result: Optional[Detections],
         exception: Exception
 ) -> None:
     with exception:
         result = detections[index]
         assert result == expected_result
+
+
+@pytest.mark.parametrize(
+    'detections_list, expected_result, exception',
+    [
+        (
+            [],
+            Detections.empty(),
+            DoesNotRaise()
+        ),  # empty detections list
+        (
+            [
+                Detections.empty()
+            ],
+            Detections.empty(),
+            DoesNotRaise()
+        ),  # single empty detections
+        (
+            [
+                mock_detections(xyxy=[[10, 10, 20, 20]])
+            ],
+            mock_detections(xyxy=[[10, 10, 20, 20]]),
+            DoesNotRaise()
+        ),  # single detection with xyxy field
+        (
+            [
+                mock_detections(xyxy=[[10, 10, 20, 20]]),
+                Detections.empty()
+            ],
+            mock_detections(xyxy=[[10, 10, 20, 20]]),
+            DoesNotRaise()
+        ),  # single detection with xyxy field + empty detection
+        (
+            [
+                mock_detections(xyxy=[[10, 10, 20, 20]]),
+                mock_detections(xyxy=[[20, 20, 30, 30]])
+            ],
+            mock_detections(
+                xyxy=[
+                    [10, 10, 20, 20],
+                    [20, 20, 30, 30]
+                ]),
+            DoesNotRaise()
+        ),  # two detections with xyxy field
+        (
+            [
+                mock_detections(
+                    xyxy=[[10, 10, 20, 20]],
+                    class_id=[0]),
+                mock_detections(
+                    xyxy=[[20, 20, 30, 30]])
+            ],
+            mock_detections(
+                xyxy=[
+                    [10, 10, 20, 20],
+                    [20, 20, 30, 30]
+                ]),
+            DoesNotRaise()
+        ),  # detection with xyxy, class_id fields + detection with xyxy field
+(
+            [
+                mock_detections(
+                    xyxy=[[10, 10, 20, 20]],
+                    class_id=[0]),
+                mock_detections(
+                    xyxy=[[20, 20, 30, 30]],
+                    class_id=[1]),
+            ],
+            mock_detections(
+                xyxy=[
+                    [10, 10, 20, 20],
+                    [20, 20, 30, 30]
+                ],
+                class_id=[0, 1]
+            ),
+            DoesNotRaise()
+        ),  # two detections with xyxy, class_id fields
+    ]
+)
+def test_merge(
+        detections_list: List[Detections],
+        expected_result: Optional[Detections],
+        exception: Exception
+) -> None:
+    with exception:
+        result = Detections.merge(detections_list=detections_list)
+        assert result == expected_result

test/detection/test_utils.py

@@ -1,11 +1,12 @@
 from contextlib import ExitStack as DoesNotRaise
-from typing import Optional, Tuple
+from typing import Optional, Tuple, List
 
 import pytest
 
 import numpy as np
 
-from supervision.detection.utils import non_max_suppression, clip_boxes
+from supervision.detection.utils import non_max_suppression, clip_boxes, filter_polygons_by_area, \
+    process_roboflow_result
 
 
 @pytest.mark.parametrize(
@@ -186,3 +187,151 @@ def test_non_max_suppression(
 def test_clip_boxes(boxes_xyxy: np.ndarray, frame_resolution_wh: Tuple[int, int], expected_result: np.ndarray) -> None:
     result = clip_boxes(boxes_xyxy=boxes_xyxy, frame_resolution_wh=frame_resolution_wh)
     assert np.array_equal(result, expected_result)
+
+
+@pytest.mark.parametrize(
+    "polygons, min_area, max_area, expected_result, exception",
+    [
+        (
+            [np.array([[0, 0], [0, 10], [10, 10], [10, 0]])],
+            None,
+            None,
+            [np.array([[0, 0], [0, 10], [10, 10], [10, 0]])],
+            DoesNotRaise()
+        ),  # single polygon without area constraints
+        (
+            [np.array([[0, 0], [0, 10], [10, 10], [10, 0]])],
+            50,
+            None,
+            [np.array([[0, 0], [0, 10], [10, 10], [10, 0]])],
+            DoesNotRaise()
+        ),  # single polygon with min_area constraint
+        (
+            [np.array([[0, 0], [0, 10], [10, 10], [10, 0]])],
+            None,
+            50,
+            [],
+            DoesNotRaise()
+        ),  # single polygon with max_area constraint
+        (
+            [
+                np.array([[0, 0], [0, 10], [10, 10], [10, 0]]),
+                np.array([[0, 0], [0, 20], [20, 20], [20, 0]])
+            ],
+            200,
+            None,
+            [np.array([[0, 0], [0, 20], [20, 20], [20, 0]])],
+            DoesNotRaise()
+        ),  # two polygons with min_area constraint
+        (
+            [
+                np.array([[0, 0], [0, 10], [10, 10], [10, 0]]),
+                np.array([[0, 0], [0, 20], [20, 20], [20, 0]])
+            ],
+            None,
+            200,
+            [np.array([[0, 0], [0, 10], [10, 10], [10, 0]])],
+            DoesNotRaise()
+        ),  # two polygons with max_area constraint
+        (
+            [
+                np.array([[0, 0], [0, 10], [10, 10], [10, 0]]),
+                np.array([[0, 0], [0, 20], [20, 20], [20, 0]])
+            ],
+            200,
+            200,
+            [],
+            DoesNotRaise()
+        ),  # two polygons with both area constraints
+        (
+            [
+                np.array([[0, 0], [0, 10], [10, 10], [10, 0]]),
+                np.array([[0, 0], [0, 20], [20, 20], [20, 0]])
+            ],
+            100,
+            100,
+            [np.array([[0, 0], [0, 10], [10, 10], [10, 0]])],
+            DoesNotRaise()
+        ),  # two polygons with min_area and max_area equal to the area of the first polygon
+        (
+            [
+                np.array([[0, 0], [0, 10], [10, 10], [10, 0]]),
+                np.array([[0, 0], [0, 20], [20, 20], [20, 0]])
+            ],
+            400,
+            400,
+            [np.array([[0, 0], [0, 20], [20, 20], [20, 0]])],
+            DoesNotRaise()
+        ),  # two polygons with min_area and max_area equal to the area of the second polygon
+    ]
+)
+def test_filter_polygons_by_area(
+        polygons: List[np.ndarray],
+        min_area: Optional[float],
+        max_area: Optional[float],
+        expected_result: List[np.ndarray],
+        exception: Exception
+) -> None:
+    with exception:
+        result = filter_polygons_by_area(polygons=polygons, min_area=min_area, max_area=max_area)
+        assert len(result) == len(expected_result)
+        for result_polygon, expected_result_polygon in zip(result, expected_result):
+            assert np.array_equal(result_polygon, expected_result_polygon)
+
+
+@pytest.mark.parametrize(
+    "roboflow_result, class_list, expected_result, exception",
+    [
+        (
+            {
+                "predictions": [],
+                "image": {"width": 1000, "height": 1000}
+            },
+            ["person", "car", "truck"],
+            (
+                np.empty((0, 4)),
+                np.empty(0),
+                np.empty(0),
+                None
+             ),
+            DoesNotRaise()
+        ),  # empty result
+        (
+            {
+                "predictions": [
+                    {
+                        "x": 200.0,
+                        "y": 300.0,
+                        "width": 50.0,
+                        "height": 50.0,
+                        "confidence": 0.9,
+                        "class": "person"
+                    }
+                ],
+                "image": {"width": 1000, "height": 1000}
+            },
+            ["person", "car", "truck"],
+            (
+                np.array([
+                    [175.0, 275.0, 225.0, 325.0]
+                ]),
+                np.array([0.9]),
+                np.array([0]),
+                None
+             ),
+            DoesNotRaise()
+        ),  # single bounding box
+    ]
+)
+def test_process_roboflow_result(
+        roboflow_result: dict,
+        class_list: List[str],
+        expected_result: Tuple[np.ndarray, np.ndarray, np.ndarray, Optional[np.ndarray]],
+        exception: Exception
+) -> None:
+    with exception:
+        result = process_roboflow_result(roboflow_result=roboflow_result, class_list=class_list)
+        assert np.array_equal(result[0], expected_result[0])
+        assert np.array_equal(result[1], expected_result[1])
+        assert np.array_equal(result[2], expected_result[2])
+        assert (result[3] is None and expected_result[3] is None) or (np.array_equal(result[3], expected_result[3]))

test/metrics/test_detection.py

@@ -0,0 +1,442 @@
+from contextlib import ExitStack as DoesNotRaise
+from typing import Optional, Union
+
+import numpy as np
+import pytest
+
+from supervision.detection.core import Detections
+from supervision.metrics.detection import ConfusionMatrix, detections_to_tensor, MeanAveragePrecision
+from test.utils import mock_detections, assert_almost_equal
+
+CLASSES = np.arange(80)
+NUM_CLASSES = len(CLASSES)
+
+PREDICTIONS = np.array(
+    [
+        [2254, 906, 2447, 1353, 0.90538, 0],
+        [2049, 1133, 2226, 1371, 0.59002, 56],
+        [727, 1224, 838, 1601, 0.51119, 39],
+        [808, 1214, 910, 1564, 0.45287, 39],
+        [6, 52, 1131, 2133, 0.45057, 72],
+        [299, 1225, 512, 1663, 0.45029, 39],
+        [529, 874, 645, 945, 0.31101, 39],
+        [8, 47, 1935, 2135, 0.28192, 72],
+        [2265, 813, 2328, 901, 0.2714, 62],
+    ],
+    dtype=np.float32,
+)
+
+TARGET_TENSORS = [
+    np.array(
+        [
+            [2254, 906, 2447, 1353, 0],
+            [2049, 1133, 2226, 1371, 56],
+            [727, 1224, 838, 1601, 39],
+            [808, 1214, 910, 1564, 39],
+            [6, 52, 1131, 2133, 72],
+            [299, 1225, 512, 1663, 39],
+            [529, 874, 645, 945, 39],
+            [8, 47, 1935, 2135, 72],
+            [2265, 813, 2328, 901, 62],
+        ]
+    )
+]
+
+DETECTIONS = Detections(
+    xyxy=PREDICTIONS[:, :4],
+    confidence=PREDICTIONS[:, 4],
+    class_id=PREDICTIONS[:, 5].astype(int),
+)
+CERTAIN_DETECTIONS = Detections(
+    xyxy=PREDICTIONS[:, :4],
+    confidence=np.ones(len(PREDICTIONS)),
+    class_id=PREDICTIONS[:, 5].astype(int),
+)
+
+DETECTION_TENSORS = [
+    np.concatenate(
+        [
+            det.xyxy,
+            np.expand_dims(det.class_id, 1),
+            np.expand_dims(det.confidence, 1),
+        ],
+        axis=1,
+    )
+    for det in [DETECTIONS]
+]
+CERTAIN_DETECTION_TENSORS = [
+    np.concatenate(
+        [
+            det.xyxy,
+            np.expand_dims(det.class_id, 1),
+            np.ones((len(det), 1)),
+        ],
+        axis=1,
+    )
+    for det in [DETECTIONS]
+]
+
+IDEAL_MATCHES = np.stack(
+    [
+        np.arange(len(PREDICTIONS)),
+        np.arange(len(PREDICTIONS)),
+        np.ones(len(PREDICTIONS)),
+    ],
+    axis=1,
+)
+
+
+def create_empty_conf_matrix(num_classes: int, do_add_dummy_class: bool = True):
+    if do_add_dummy_class:
+        num_classes += 1
+    return np.zeros((num_classes, num_classes))
+
+
+def update_ideal_conf_matrix(conf_matrix: np.ndarray, class_ids: np.ndarray):
+    for class_id, count in zip(*np.unique(class_ids, return_counts=True)):
+        class_id = int(class_id)
+        conf_matrix[class_id, class_id] += count
+    return conf_matrix
+
+
+def worsen_ideal_conf_matrix(
+    conf_matrix: np.ndarray, class_ids: Union[np.ndarray, list]
+):
+    for class_id in class_ids:
+        class_id = int(class_id)
+        conf_matrix[class_id, class_id] -= 1
+        conf_matrix[class_id, 80] += 1
+    return conf_matrix
+
+
+IDEAL_CONF_MATRIX = create_empty_conf_matrix(NUM_CLASSES)
+IDEAL_CONF_MATRIX = update_ideal_conf_matrix(IDEAL_CONF_MATRIX, PREDICTIONS[:, 5])
+
+GOOD_CONF_MATRIX = worsen_ideal_conf_matrix(IDEAL_CONF_MATRIX.copy(), [62, 72])
+
+BAD_CONF_MATRIX = worsen_ideal_conf_matrix(
+    IDEAL_CONF_MATRIX.copy(), [62, 72, 72, 39, 39, 39, 39, 56]
+)
+
+
+@pytest.mark.parametrize(
+    "detections, with_confidence, expected_result, exception",
+    [
+        (
+            Detections.empty(),
+            False,
+            np.empty((0, 5), dtype=np.float32),
+            DoesNotRaise(),
+        ),  # empty detections; no confidence
+        (
+            Detections.empty(),
+            True,
+            np.empty((0, 6), dtype=np.float32),
+            DoesNotRaise(),
+        ),  # empty detections; with confidence
+        (
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0], confidence=[0.5]),
+            False,
+            np.array([[0, 0, 10, 10, 0]], dtype=np.float32),
+            DoesNotRaise(),
+        ),  # single detection; no confidence
+        (
+            mock_detections(xyxy=[[0, 0, 10, 10]], class_id=[0], confidence=[0.5]),
+            True,
+            np.array([[0, 0, 10, 10, 0, 0.5]], dtype=np.float32),
+            DoesNotRaise(),
+        ),  # single detection; with confidence
+        (
+            mock_detections(xyxy=[[0, 0, 10, 10], [0, 0, 20, 20]], class_id=[0, 1], confidence=[0.5, 0.2]),
+            False,
+            np.array([[0, 0, 10, 10, 0], [0, 0, 20, 20, 1]], dtype=np.float32),
+            DoesNotRaise(),
+        ),  # multiple detections; no confidence
+        (
+            mock_detections(xyxy=[[0, 0, 10, 10], [0, 0, 20, 20]], class_id=[0, 1], confidence=[0.5, 0.2]),
+            True,
+            np.array([[0, 0, 10, 10, 0, 0.5], [0, 0, 20, 20, 1, 0.2]], dtype=np.float32),
+            DoesNotRaise(),
+        ),  # multiple detections; with confidence
+    ],
+)
+def test_detections_to_tensor(
+    detections: Detections,
+    with_confidence: bool,
+    expected_result: Optional[np.ndarray],
+    exception: Exception
+):
+    with exception:
+        result = detections_to_tensor(
+            detections=detections,
+            with_confidence=with_confidence
+        )
+        assert np.array_equal(result, expected_result)
+
+
+@pytest.mark.parametrize(
+    "predictions, targets, classes, conf_threshold, iou_threshold, expected_result, exception",
+    [
+        (
+            DETECTION_TENSORS,
+            TARGET_TENSORS,
+            CLASSES,
+            0.2,
+            0.5,
+            IDEAL_CONF_MATRIX,
+            DoesNotRaise(),
+        ),
+        (
+            [],
+            [],
+            CLASSES,
+            0.2,
+            0.5,
+            create_empty_conf_matrix(NUM_CLASSES),
+            DoesNotRaise(),
+        ),
+        (
+            DETECTION_TENSORS,
+            TARGET_TENSORS,
+            CLASSES,
+            0.3,
+            0.5,
+            GOOD_CONF_MATRIX,
+            DoesNotRaise(),
+        ),
+        (
+            DETECTION_TENSORS,
+            TARGET_TENSORS,
+            CLASSES,
+            0.6,
+            0.5,
+            BAD_CONF_MATRIX,
+            DoesNotRaise(),
+        ),
+        (
+            [
+                np.array(
+                    [
+                        [0.0, 0.0, 3.0, 3.0, 0, 0.9],  # correct detection of [0]
+                        [
+                            0.1,
+                            0.1,
+                            3.0,
+                            3.0,
+                            0,
+                            0.9,
+                        ],  # additional detection of [0] - FP
+                        [
+                            6.0,
+                            1.0,
+                            8.0,
+                            3.0,
+                            1,
+                            0.8,
+                        ],  # correct detection with incorrect class
+                        [1.0, 6.0, 2.0, 7.0, 1, 0.8],  # incorrect detection - FP
+                        [
+                            1.0,
+                            2.0,
+                            2.0,
+                            4.0,
+                            1,
+                            0.8,
+                        ],  # incorrect detection with low IoU - FP
+                    ]
+                )
+            ],
+            [
+                np.array(
+                    [
+                        [0.0, 0.0, 3.0, 3.0, 0],  # [0] detected
+                        [2.0, 2.0, 5.0, 5.0, 1],  # [1] undetected - FN
+                        [
+                            6.0,
+                            1.0,
+                            8.0,
+                            3.0,
+                            2,
+                        ],  # [2] correct detection with incorrect class
+                    ]
+                )
+            ],
+            CLASSES[:3],
+            0.6,
+            0.5,
+            np.array([[1, 0, 0, 0], [0, 0, 0, 1], [0, 1, 0, 0], [1, 2, 0, 0]]),
+            DoesNotRaise(),
+        ),
+        (
+            [
+                np.array(
+                    [
+                        [0.0, 0.0, 3.0, 3.0, 0, 0.9],  # correct detection of [0]
+                        [
+                            0.1,
+                            0.1,
+                            3.0,
+                            3.0,
+                            0,
+                            0.9,
+                        ],  # additional detection of [0] - FP
+                        [
+                            6.0,
+                            1.0,
+                            8.0,
+                            3.0,
+                            1,
+                            0.8,
+                        ],  # correct detection with incorrect class
+                        [1.0, 6.0, 2.0, 7.0, 1, 0.8],  # incorrect detection - FP
+                        [
+                            1.0,
+                            2.0,
+                            2.0,
+                            4.0,
+                            1,
+                            0.8,
+                        ],  # incorrect detection with low IoU - FP
+                    ]
+                )
+            ],
+            [
+                np.array(
+                    [
+                        [0.0, 0.0, 3.0, 3.0, 0],  # [0] detected
+                        [2.0, 2.0, 5.0, 5.0, 1],  # [1] undetected - FN
+                        [
+                            6.0,
+                            1.0,
+                            8.0,
+                            3.0,
+                            2,
+                        ],  # [2] correct detection with incorrect class
+                    ]
+                )
+            ],
+            CLASSES[:3],
+            0.6,
+            1.0,
+            np.array([[0, 0, 0, 1], [0, 0, 0, 1], [0, 0, 0, 1], [2, 3, 0, 0]]),
+            DoesNotRaise(),
+        ),
+    ],
+)
+def test_from_tensors(
+    predictions,
+    targets,
+    classes,
+    conf_threshold,
+    iou_threshold,
+    expected_result: Optional[np.ndarray],
+    exception: Exception,
+):
+    with exception:
+        result = ConfusionMatrix.from_tensors(
+            predictions=predictions,
+            targets=targets,
+            classes=classes,
+            conf_threshold=conf_threshold,
+            iou_threshold=iou_threshold,
+        )
+
+        assert result.matrix.diagonal().sum() == expected_result.diagonal().sum()
+        assert np.array_equal(result.matrix, expected_result)
+
+
+@pytest.mark.parametrize(
+    "predictions, targets, num_classes, conf_threshold, iou_threshold, expected_result, exception",
+    [
+        (
+            DETECTION_TENSORS[0],
+            CERTAIN_DETECTION_TENSORS[0],
+            NUM_CLASSES,
+            0.2,
+            0.5,
+            IDEAL_CONF_MATRIX,
+            DoesNotRaise(),
+        )
+    ],
+)
+def test_evaluate_detection_batch(
+    predictions,
+    targets,
+    num_classes,
+    conf_threshold,
+    iou_threshold,
+    expected_result: Optional[np.ndarray],
+    exception: Exception,
+):
+    with exception:
+        result = ConfusionMatrix.evaluate_detection_batch(
+            predictions=predictions,
+            targets=targets,
+            num_classes=num_classes,
+            conf_threshold=conf_threshold,
+            iou_threshold=iou_threshold,
+        )
+
+        assert result.diagonal().sum() == result.sum()
+        assert np.array_equal(result, expected_result)
+
+
+@pytest.mark.parametrize(
+    "matches, expected_result, exception",
+    [
+        (
+            IDEAL_MATCHES,
+            IDEAL_MATCHES,
+            DoesNotRaise(),
+        )
+    ],
+)
+def test_drop_extra_matches(
+    matches,
+    expected_result: Optional[np.ndarray],
+    exception: Exception,
+):
+    with exception:
+        result = ConfusionMatrix._drop_extra_matches(matches)
+
+        assert np.array_equal(result, expected_result)
+
+
+@pytest.mark.parametrize(
+    'recall, precision, expected_result, exception',
+    [
+        (
+            np.array([1.0]),
+            np.array([1.0]),
+            1.0,
+            DoesNotRaise()
+        ),  # perfect recall and precision
+        (
+            np.array([0.0]),
+            np.array([0.0]),
+            0.0,
+            DoesNotRaise()
+        ),  # no recall and precision
+        (
+            np.array([0.0, 0.2, 0.2, 0.8, 0.8, 1.0]),
+            np.array([0.7, 0.8, 0.4, 0.5, 0.1, 0.2]),
+            0.5,
+            DoesNotRaise()
+        ),
+        (
+            np.array([0.0, 0.5, 0.5, 1.0]),
+            np.array([0.75, 0.75, 0.75, 0.75]),
+            0.75,
+            DoesNotRaise()
+        )
+    ]
+)
+def test_compute_average_precision(
+        recall: np.ndarray,
+        precision: np.ndarray,
+        expected_result: float,
+        exception: Exception
+) -> None:
+    with exception:
+        result = MeanAveragePrecision.compute_average_precision(recall=recall, precision=precision)
+        assert_almost_equal(result, expected_result, tolerance=0.01)

test/utils.py

@@ -0,0 +1,27 @@
+from typing import List
+
+import numpy as np
+
+from supervision.detection.core import Detections
+
+
+def mock_detections(
+    xyxy: List[List[float]],
+    confidence: List[float] = None,
+    class_id: List[int] = None,
+    tracker_id: List[int] = None,
+) -> Detections:
+    return Detections(
+        xyxy=np.array(xyxy, dtype=np.float32),
+        confidence=confidence
+        if confidence is None
+        else np.array(confidence, dtype=np.float32),
+        class_id=class_id if class_id is None else np.array(class_id, dtype=int),
+        tracker_id=tracker_id
+        if tracker_id is None
+        else np.array(tracker_id, dtype=int),
+    )
+
+
+def assert_almost_equal(actual, expected, tolerance=1e-5):
+    assert abs(actual - expected) < tolerance, f"Expected {expected}, but got {actual}."