Abstract
Accurate hand motion capture and standardized 3D representation are essentialfor various hand-related tasks. Collecting keypoints-only data, while efficientand cost-effective, results in low-fidelity representations and lacks surfaceinformation. Furthermore, data inconsistencies across sources challenge theirintegration and use. We present UniHands, a novel method for creatingstandardized yet personalized hand models from wild-collected keypoints fromdiverse sources. Unlike existing neural implicit representation methods,UniHands uses the widely-adopted parametric models MANO and NIMBLE, providing amore scalable and versatile solution. It also derives unified hand joints fromthe meshes, which facilitates seamless integration into various hand-relatedtasks. Experiments on the FreiHAND and InterHand2.6M datasets demonstrate itsability to precisely reconstruct hand mesh vertices and keypoints, effectivelycapturing high-degree articulation motions. Empirical studies involving nineparticipants show a clear preference for our unified joints over existingconfigurations for accuracy and naturalism (p-value 0.016).