MoCapAnything V2: End-to-End Motion Capture for Arbitrary Skeletons

MoCapAnything V2 is a fully end-to-end motion capture framework for arbitrary skeletons from monocular video.

It jointly predicts joint positions and joint rotations, using reference pose-rotation pairs to reduce rotation ambiguity.

The system reports lower rotation error and faster inference than mesh-based baselines.

By learning rotations directly, it improves accuracy on unseen skeletons and removes a key limitation of earlier pipelines.