Convert landmarks to Human22 schema with COCO or Halpe mode

Use This When

• Standardizing pose landmarks to 22-keypoint Human22 schema for body analysis
• Converting COCO-17 or Halpe-26 landmarks to unified 22-point representation
• Building pose pipelines requiring extended body landmarks beyond basic COCO-17
• Interfacing with components expecting neck, thorax, spine, hip, and head keypoints

What It Does

• Converts landmark lists to Human22 schema with 22 named body keypoints
• COCO mode: interpolates missing keypoints (neck, thorax, spine, hip) from COCO-17 landmarks
• Halpe mode: uses native Halpe keypoints for head, neck, hip with computed thorax/spine
• Estimates head position using shoulder/ear/eye geometry when not directly available

Works Best With

• detect-landmarks → this component → detect-bodypart for extended body part extraction
• Pose estimation → this component → pose analysis requiring torso or head landmarks
• Integration with filter-pose-bbox or orientation-based filtering
• Feeding visualization components for skeleton rendering with extended keypoints

Caveats

• COCO mode requires minimum 17 landmarks; Halpe mode requires minimum 26 landmarks
• Interpolated keypoints have confidence based on source points; may be lower than direct detections
• Head estimation uses heuristics calibrated on COCO dataset; accuracy varies by pose
• Mode selection must match upstream pose model output; verify landmark count compatibility