JoHan is a unified generative framework that recovers 2D and 3D hand poses from video without intermediate per-frame predictions. It learns temporal dynamics and cross-representation correspondence to ensure temporal consistency despite occlusions.
HOW THIS AFFECTS YOU
●
builderThis offers a more robust method for 3D hand tracking in video-based applications.
●
researcherThis avoids the error accumulation common in frame-wise pose estimation pipelines.