Fix lossy COCO round-trip: preserve area and iscrowd in as_coco

src/supervision/dataset/formats/coco.py

@@ -164,14 +164,18 @@ def detections_to_coco_annotations(
     approximation_percentage: float = 0.75,
 ) -> tuple[list[CocoDict], int]:
     coco_annotations: list[CocoDict] = []
-    for xyxy, mask, _, class_id, _, _ in detections:
+    for xyxy, mask, _, class_id, _, data in detections:
         if class_id is None:
             raise ValueError("Detections must include class_id for COCO export.")
         box_width, box_height = xyxy[2] - xyxy[0], xyxy[3] - xyxy[1]
         segmentation: Union[list[list[float]], dict[str, list[int]]] = []
-        iscrowd = 0
         if mask is not None:
-            iscrowd = contains_holes(mask=mask) or contains_multiple_segments(mask=mask)
+            if "iscrowd" in data:
+                iscrowd = int(np.asarray(data["iscrowd"]).item())
+            else:
+                iscrowd = int(
+                    contains_holes(mask=mask) or contains_multiple_segments(mask=mask)
+                )
 
             if iscrowd:
                 segmentation = {
@@ -196,12 +200,16 @@ def detections_to_coco_annotations(
                         "returned no polygons.",
                         stacklevel=2,
                     )
+        else:
+            iscrowd = int(np.asarray(data.get("iscrowd", 0)).item())
+
+        area: float = float(np.asarray(data.get("area", box_width * box_height)).item())
         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,
+            "area": area,
             "segmentation": segmentation,
             "iscrowd": iscrowd,
         }

tests/dataset/formats/test_coco.py

@@ -947,6 +947,110 @@ def test_detections_to_coco_annotations_handles_empty_approximated_polygons() ->
     assert annotations[0]["iscrowd"] == 0
 
 
+def test_detections_to_coco_annotations_preserves_area_from_data() -> None:
+    """area stored in detections.data should be used instead of bbox area."""
+    detections = Detections(
+        xyxy=np.array([[10.0, 20.0, 110.0, 120.0]], dtype=np.float32),
+        class_id=np.array([0], dtype=int),
+        data={"iscrowd": np.array([0], dtype=int), "area": np.array([5000.0])},
+    )
+
+    annotations, _ = detections_to_coco_annotations(
+        detections=detections,
+        image_id=1,
+        annotation_id=1,
+    )
+
+    assert len(annotations) == 1
+    assert annotations[0]["area"] == 5000.0
+    assert annotations[0]["iscrowd"] == 0
+    assert type(annotations[0]["iscrowd"]) is int
+
+
+def test_detections_to_coco_annotations_preserves_iscrowd_from_data_when_no_mask() -> (
+    None
+):
+    """iscrowd stored in detections.data should be used when no mask is present."""
+    detections = Detections(
+        xyxy=np.array([[0.0, 0.0, 100.0, 100.0]], dtype=np.float32),
+        class_id=np.array([0], dtype=int),
+        data={"iscrowd": np.array([1], dtype=int), "area": np.array([1234.5])},
+    )
+
+    annotations, _ = detections_to_coco_annotations(
+        detections=detections,
+        image_id=1,
+        annotation_id=1,
+    )
+
+    assert len(annotations) == 1
+    assert annotations[0]["iscrowd"] == 1
+    assert type(annotations[0]["iscrowd"]) is int
+    assert annotations[0]["area"] == 1234.5
+
+
+def test_detections_to_coco_annotations_iscrowd_is_int_when_mask_provided() -> None:
+    """iscrowd should be stored as int (0 or 1), not as Python bool."""
+    mask = np.zeros((1, 5, 5), dtype=bool)
+    mask[0, 0:3, 0:3] = True  # simple single-component rectangle
+
+    detections = Detections(
+        xyxy=np.array([[0.0, 0.0, 3.0, 3.0]], dtype=np.float32),
+        class_id=np.array([0], dtype=int),
+        mask=mask,
+    )
+
+    annotations, _ = detections_to_coco_annotations(
+        detections=detections,
+        image_id=1,
+        annotation_id=1,
+    )
+
+    assert len(annotations) == 1
+    assert annotations[0]["iscrowd"] == 0
+    assert type(annotations[0]["iscrowd"]) is int
+
+
+def test_detections_to_coco_annotations_data_area_overrides_bbox_with_mask() -> None:
+    """data["area"] should override computed bbox area even when a mask is present."""
+    mask = np.zeros((1, 10, 10), dtype=bool)
+    mask[0, 0:4, 0:4] = True  # 16-pixel polygon area
+
+    detections = Detections(
+        xyxy=np.array([[0.0, 0.0, 10.0, 10.0]], dtype=np.float32),
+        class_id=np.array([0], dtype=int),
+        mask=mask,
+        data={"area": np.array([999.0])},
+    )
+
+    annotations, _ = detections_to_coco_annotations(
+        detections=detections,
+        image_id=1,
+        annotation_id=1,
+    )
+
+    assert len(annotations) == 1
+    assert annotations[0]["area"] == 999.0
+
+
+def test_detections_to_coco_annotations_fallback_area_when_no_data() -> None:
+    """When detections have no area in data, area should fall back to bbox area."""
+    detections = Detections(
+        xyxy=np.array([[10.0, 20.0, 110.0, 120.0]], dtype=np.float32),
+        class_id=np.array([0], dtype=int),
+    )
+
+    annotations, _ = detections_to_coco_annotations(
+        detections=detections,
+        image_id=1,
+        annotation_id=1,
+    )
+
+    assert len(annotations) == 1
+    assert annotations[0]["area"] == 100.0 * 100.0
+    assert annotations[0]["iscrowd"] == 0
+
+
 def test_load_coco_annotations_infers_masks_from_segmentation_field(
     tmp_path, coco_data_with_and_without_segmentation: dict[str, object]
 ) -> None: