Locate Bodypart

1 VersionDetect

Crops named body regions out of an Image guided by one or more [Landmarks2d.Human22] poses, emitting a per-person [(Image, Double)] of region image plus a confidence so a downstream classifier can work on the isolated part rather than the whole frame.

How it fits

Typical backends

A pose detector's keypoints are upgraded to the 22-joint schema by label_coco_mpii_body_landmarks, the region is cropped, and the result is drawn back onto the frame by visualize.

The same crop drives a per-person classifier that consumes the isolated region image and emits a label, decoupling recognition from frame layout.

Caveats

Limitationtargets accepts helmet, head, eyes, mouth, torso, lArm, rArm, lForearm, rForearm, lHand, rHand, lLeg, rLeg, lLowerLeg, rLowerLeg, lFoot and rFoot. Arms and legs are returned as masked limb strokes, the rest as cropped rectangles. An unrecognised name aborts the component at startup before it runs.
AccuracyThe eyes crop is mirrored automatically when the subject faces their own left, so the region reads consistently regardless of head orientation.
I/O contractThe emitted confidence on output is the lowest of the combined parts, so a composite region inherits its weakest constituent's score.
Fallbackmin_confidence always keeps the single strongest part even below threshold, so a valid pose never yields an empty crop; gate on the returned Double downstream to actually drop weak results.
Hard constraintThe pose input must be the 22-keypoint Landmarks2d.Human22 schema; other keypoint conventions need a converter ahead of this component.

Versions

381dca09latestdefault
Automated release
6/24/2026