Skeleton

Skeleton annotation extends keypoint annotation by connecting keypoints with edges to form a graph that represents the topological structure of an object, most often a human or animal.

Each edge is typically called a "limb” or “bone” and captures the spatial relationship between two keypoints, such as the link from shoulder to elbow.

Skeleton data supports pose estimation, action recognition, motion capture, and animation.

Compared with isolated keypoints, a skeleton provides richer structural priors, helping models learn more plausible spatial relationships and motion constraints.