detection_map

Formula

detection_map computes average precision per label and returns the mean across labels. A prediction matches a ground-truth box when IoU is at least 0.5.

Hand-Computable Example

One predicted box with bounding box [0, 0, 2, 2] and one ground-truth box with the same bounding box.

Expected output: 1.0.

Edge Cases

  • Missing detection arrays return 0.0.
  • No ground-truth boxes returns 1.0 only when there are also no predictions.
  • Invalid boxes raise an error.

Limitations

This is a compact deterministic implementation for local regression checks. It is not a replacement for full COCO-style evaluation tooling.

Test References

  • tests/golden/test_openloop_metrics_golden.py
  • tests/test_openloop_metrics.py