fix(detection): scale `from_tensorflow` boxes by correct axes (#2360)

`Detections.from_tensorflow` scaled the normalized box coordinates by the
wrong image dimensions: the y coordinates (ymin/ymax, columns 0 and 2) were
multiplied by width and the x coordinates (xmin/xmax, columns 1 and 3) by
height. Tensorflow Hub object-detection models emit `detection_boxes` as
normalized `[ymin, xmin, ymax, xmax]`, so y must scale by height and x by
width.

The bug is masked on square images (width == height) but corrupts every
coordinate on the common non-square case — e.g. a box normalized to
`[0.1, 0.2, 0.5, 0.6]` on a 1000x500 image came out as
`[100, 100, 300, 500]` instead of the correct `[200, 50, 600, 250]`.

Swap the two multipliers so y scales by `resolution_wh[1]` (height) and x by
`resolution_wh[0]` (width). Adds a non-square regression test (the connector
was previously untested).

- Expand tensorflow_results arg to document required dict keys and tensor
  shapes so callers know what to pass before getting a KeyError
- Add Note: section documenting the [ymin, xmin, ymax, xmax] normalized
  box format; the inline comment was only visible to code readers
- Fix SOURCE_IMAGE_PATH undefined identifier → "<SOURCE_IMAGE_PATH>"
  string placeholder (consistent with other connector examples in file)

---------

Co-authored-by: Jirka Borovec <6035284+Borda@users.noreply.github.com>
Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com>

src/supervision/detection/core.py

@@ -407,7 +407,12 @@ class Detections:
         inference result.
 
         Args:
-            tensorflow_results: The output results from Tensorflow Hub.
+            tensorflow_results: Raw output dict from a TensorFlow Hub
+                object-detection model. Must contain:
+                ``"detection_boxes"`` (shape ``[1, N, 4]``, normalized
+                ``[ymin, xmin, ymax, xmax]``), ``"detection_scores"``
+                (shape ``[1, N]``), and ``"detection_classes"``
+                (shape ``[1, N]``).
             resolution_wh: The input image resolution as `(width, height)`.
                 Bounding boxes from Tensorflow are normalized and are scaled
                 to absolute coordinates using this resolution.
@@ -415,6 +420,13 @@ class Detections:
         Returns:
             A new Detections object.
 
+        Note:
+            TensorFlow Hub object-detection models return bounding boxes
+            normalized as ``[ymin, xmin, ymax, xmax]``. This method rescales
+            them to absolute pixel coordinates and reorders them to ``xyxy``
+            (``[xmin, ymin, xmax, ymax]``) before constructing the
+            :class:`Detections` object.
+
         Example:
             ```python
             import tensorflow as tf
@@ -424,7 +436,7 @@ class Detections:
 
             module_handle = "https://tfhub.dev/tensorflow/centernet/hourglass_512x512_kpts/1"
             model = hub.load(module_handle)
-            img = np.array(cv2.imread(SOURCE_IMAGE_PATH))
+            img = np.array(cv2.imread("<SOURCE_IMAGE_PATH>"))
             result = model(img)
             detections = sv.Detections.from_tensorflow(
                 result, resolution_wh=(img.shape[1], img.shape[0])
@@ -432,9 +444,11 @@ class Detections:
             ```
         """
 
+        # Tensorflow returns normalized boxes as [ymin, xmin, ymax, xmax], so the
+        # y coordinates (cols 0, 2) scale by height and x (cols 1, 3) by width.
         boxes = tensorflow_results["detection_boxes"][0].numpy()
-        boxes[:, [0, 2]] *= resolution_wh[0]
-        boxes[:, [1, 3]] *= resolution_wh[1]
+        boxes[:, [0, 2]] *= resolution_wh[1]
+        boxes[:, [1, 3]] *= resolution_wh[0]
         boxes = boxes[:, [1, 0, 3, 2]]
         return cls(
             xyxy=boxes,

tests/detection/test_from_adapters.py

@@ -5,6 +5,7 @@ import supervision.detection.core as detection_core
 from supervision.config import CLASS_NAME_DATA_FIELD
 from supervision.detection.core import Detections
 from tests.helpers import (
+    _FakeTensor,
     _FakeUltralyticsBoxes,
     _FakeUltralyticsResults,
     _FakeYoloNasPrediction,
@@ -109,3 +110,21 @@ def test_from_yolo_nas_handles_empty_and_non_empty(
         np.testing.assert_allclose(det.xyxy, bboxes)
         np.testing.assert_allclose(det.confidence, conf)
         np.testing.assert_array_equal(det.class_id, labels.astype(int))
+
+
+def test_from_tensorflow_scales_axes_on_non_square_image() -> None:
+    """Non-square image exposes swapped scaling: y uses height, x uses width."""
+    results = {
+        "detection_boxes": [
+            _FakeTensor(np.array([[0.1, 0.2, 0.5, 0.6]], dtype=np.float32))
+        ],
+        "detection_scores": [_FakeTensor(np.array([0.9], dtype=np.float32))],
+        "detection_classes": [_FakeTensor(np.array([1], dtype=np.float32))],
+    }
+
+    det = Detections.from_tensorflow(results, resolution_wh=(1000, 500))
+
+    # xmin=0.2*1000, ymin=0.1*500, xmax=0.6*1000, ymax=0.5*500
+    np.testing.assert_allclose(det.xyxy, [[200.0, 50.0, 600.0, 250.0]])
+    np.testing.assert_allclose(det.confidence, [0.9])
+    np.testing.assert_array_equal(det.class_id, [1])