Keypoints

Keypoint annotation marks a set of predefined locations on an object, such as facial landmarks, human joints, vehicle corners, or anatomical reference points.

Each keypoint typically records 2D or 3D coordinates along with a visibility attribute that indicates whether the point is occluded or outside the frame.

This annotation type supports tasks such as human pose estimation, face recognition, gesture analysis, and sports motion analysis.

Widely used datasets include COCO Keypoints and MPII Human Pose, which define consistent keypoint schemas for training and evaluation.