نسخة أولية وصول مفتوح
Video-Conditioned Generative Joint 2D-3D Hand Motion Recovery
Recovering faithful 3D hand motion from video remains challenging due to frequent occlusions and incomplete visual observations, which make frame-wise pose estimates unreliable and temporally inconsistent. To address this problem, we propose JoHan, a unified generative framework that recovers hand motion directly from …