roboflow/supervision · error · ValueError

OBB data for each detection must have shape (4, 2), got {cor

Error message

OBB data for each detection must have shape (4, 2), got {corners.shape}. Ensure `detections.data['{ORIENTED_BOX_COORDINATES}']` has shape (N, 4, 2) before exporting.

What it means

Raised during OBB YOLO export when the per-detection oriented corner array in data[ORIENTED_BOX_COORDINATES] does not have shape (4, 2). Each oriented box is defined by exactly 4 corner points with x,y coordinates; a wrong shape means the stored geometry is not a valid single OBB (e.g. a flattened array, wrong corner count, or the whole (N, 4, 2) batch stored per detection).

Source

Thrown at src/supervision/dataset/formats/yolo.py:398

            UserWarning,
            stacklevel=2,
        )

    annotation: list[str] = []
    for xyxy, mask, _, class_id, _, data in detections:
        if class_id is None:
            raise ValueError("Class ID is required for YOLO annotations.")
        if not isinstance(class_id, (int, np.integer)):
            raise ValueError(
                f"Detections class_id must be an integer for YOLO export, "
                f"got {type(class_id)!r}."
            )
        class_id_int = int(class_id)

        if is_obb:
            corners = np.asarray(data[ORIENTED_BOX_COORDINATES], dtype=np.float32)
            if corners.shape != (4, 2):
                raise ValueError(
                    f"OBB data for each detection must have shape (4, 2), "
                    f"got {corners.shape}. Ensure "
                    f"`detections.data['{ORIENTED_BOX_COORDINATES}']` has "
                    "shape (N, 4, 2) before exporting."
                )
            next_object = object_to_yolo(
                xyxy=xyxy,
                class_id=class_id_int,
                image_shape=image_shape,
                polygon=corners,
            )
            annotation.append(next_object)
            continue

        if mask is not None:
            polygons = approximate_mask_with_polygons(
                mask=mask,
                min_image_area_percentage=min_image_area_percentage,

View on GitHub (pinned to 7f254d9784)

Solutions

1. Reshape each detection's corners to exactly 4 rows of (x, y): corners.reshape(4, 2).
2. If starting from rotated-rect parameters, convert with cv2.boxPoints(((cx,cy),(w,h),angle)) which returns (4, 2).
3. Store the batch as one (N, 4, 2) array in detections.data; the loader indexes rows per detection.
4. Print corners.shape in a small repro to confirm the fix.

Example fix

# before
dets.data[sv.ORIENTED_BOX_COORDINATES] = flat_corners  # shape (N, 8)
# after
import numpy as np
N = len(dets)
dets.data[sv.ORIENTED_BOX_COORDINATES] = flat_corners.reshape(N, 4, 2)

import numpy as np
import supervision as sv

def obb_shapes_valid(detections) -> bool:
    """Check ORIENTED_BOX_COORDINATES is (N, 4, 2) with N matching detections."""
    c = detections.data.get(sv.ORIENTED_BOX_COORDINATES)
    return c is not None and c.shape == (len(detections), 4, 2)

try:
    lines = sv.detections_to_yolo_annotations(dets, image_shape=shape, is_obb=True)
except ValueError as e:
    if 'must have shape (4, 2)' in str(e):
        c = dets.data[sv.ORIENTED_BOX_COORDINATES]
        dets.data[sv.ORIENTED_BOX_COORDINATES] = np.asarray(c).reshape(len(dets), 4, 2)
        lines = sv.detections_to_yolo_annotations(dets, image_shape=shape, is_obb=True)
    else:
        raise

When it happens

Trigger: sv.detections_to_yolo_annotations(..., is_obb=True) where detections.data[ORIENTED_BOX_COORDINATES] was set manually to e.g. an (8,) flat array, an (N, 4, 2) array incorrectly indexed, or an (4, 3) array.

Common situations: Hand-building OBB data from model output that returns flat 8-value OBB params (x,y,w,h,angle) instead of corners — forgetting cv2.boxPoints; reshaping errors; passing per-detection rows of a batch array without indexing.

Related errors

• `is_obb=True` requires `'{ORIENTED_BOX_COORDINATES}'` in `de
• Class ID is required for YOLO annotations.
• Detections class_id must be an integer for YOLO export, got
• Cannot export dataset: image paths {first_path} and {image_p
• Expected 'names' to be a list or dict in data.yaml at '{file

AI-assisted analysis of roboflow/supervision@7f254d9784 (2026-08-15). Data as JSON: /api/errors/b68e6553b31aa5ed. Report an issue: GitHub.