Label COCO MPII Body Landmarks

@bleenco1 VersionLabel / Schema

Reindexes a flat [Landmark] body keypoint list into the 22-joint Landmarks2d.Human22 schema, interpolating neck, thorax, spine, and (COCO mode) head so COCO-17 or Halpe-26 detectors can both feed Human22 consumers.

How it fits

Typical backends

A pose estimator's keypoints are upgraded to the 22-joint schema, then drawn back onto the frame by visualize.

The Human22 record feeds filter_pose_bounding_box, whose yaw gate is built around exactly this schema's shoulder and spine joints.

Caveats

Hard constraintSource convention is fixed at startup: pose_type accepts only COCO or HALPE, and any other value aborts the component before it runs.
Hard constraintCOCO mode needs at least 17 points on input; Halpe mode needs at least 26. A shorter list violates the input contract and aborts the component, so the upstream detector must always emit a full keypoint list.
AccuracyThe neck, thorax, spine, and (COCO-mode) head joints are not detected — they are computed by interpolating real keypoints with fixed weights, so their placement and confidence are estimates that degrade when the source points are occluded.
OrderingMapping is positional and must match the convention chosen by pose_type; a list in the wrong order runs without error but binds points to the wrong named joints.

Versions

bc0a235blatestdefault
Automated release
6/25/2026