Label COCO Body Landmarks

Schema converter: takes a flat list of landmarks and emits a typed COCO Human17 record with named fields (nose, lEye, rEye, ear, shoulders, elbows, wrists, hips, knees, ankles).

How it fits

[Landmark] (17+ ordered points)  ──►  label_coco_body_landmarks  ──► Landmarks2d.Human17

Typical pipelines

• Pose analytics: detect_landmarks → label_coco_body_landmarks → calculate_landmark_angle → analytics.
• Frontal-only filtering: detect_landmarks → label_coco_body_landmarks → filter_pose_bbox → downstream.
• Pose visualisation: detect_landmarks → label_coco_body_landmarks → visualize_pose_landmarks → output_browser_stream.

Caveats

• Hard requirement: at least 17 input landmarks. Fewer triggers a StateException.
• The mapping is positional. Wrong landmark order won't error but the output names will be assigned to the wrong points.
• Indices past 16 are silently ignored.

Related components

• label_animal_ap_10k_landmarks, label_coco_mpii_body_landmarks, label_coco_wholebody_hand_landmarks, label_halpe_body_landmarks — sibling schema converters for other landmark formats.
• detect_landmarks, detect_landmarks_huggingface — typical upstream sources.
• calculate_landmark_angle, filter_pose_bbox, visualize_pose_landmarks — typical downstream consumers.