docs(changelog): add LabelMe dataset format entry (#2299)

Author: madhavcodez

docs/changelog.md

@@ -7,6 +7,8 @@ date_modified: 2026-04-30
 
 ### UnReleased
 
+- Added [#2299](https://github.com/roboflow/supervision/pull/2299): [`DetectionDataset.from_labelme`](https://supervision.roboflow.com/latest/datasets/core/#supervision.dataset.core.DetectionDataset.from_labelme) and [`DetectionDataset.as_labelme`](https://supervision.roboflow.com/latest/datasets/core/#supervision.dataset.core.DetectionDataset.as_labelme) for loading and exporting [LabelMe](https://github.com/wkentaro/labelme) per-image JSON annotations, following the existing COCO/YOLO/VOC convention. `rectangle` shapes load as boxes and `polygon` shapes as masks; unsupported shape types are skipped with a warning. The mask round-trip is a polygon approximation, not bit-exact.
+
 - Deprecated: public `validate_*` helper functions now delegate to private `_validate_*` implementations via `pydeprecate` shims. Existing imports continue to work with a `FutureWarning`; internal Supervision code uses the private helpers directly.
 
 - Fixed [#2282](https://github.com/roboflow/supervision/pull/2282): [`oriented_box_iou_batch`](https://supervision.roboflow.com/latest/detection/utils/iou_and_nms/) now allocates the rasterization canvas with correct dimensions. Previously x-extents were mapped to canvas height and y-extents to width, causing IoU to be computed on a transposed canvas for non-square oriented bounding boxes. Metrics using OBB IoU (Precision, Recall, F1Score, MeanAverageRecall with `metric_target=MetricTarget.ORIENTED_BOUNDING_BOXES`) now return correct scores.