3D Interacting Hand Pose Estimation by Hand De-occlusion and Removal

Estimating 3D interacting hand pose from a single RGB image is essential for understanding human actions. Unlike most previous works that directly predict the poses of two hands simultaneously, we propose to decompose challenging estimation task and estimate each separately. In this way, it straightforward take advantage latest research progress on single-hand system. However, in scenarios very challenging, due (1) severe hand-hand occlusion (2) ambiguity caused by homogeneous appearance hands. To tackle these challenges, novel Hand De-occlusion Removal (HDR) framework perform de-occlusion distractor removal. We also first large-scale synthetic amodal dataset, termed Amodal InterHand Dataset (AIH), facilitate model training promote development related research. Experiments show proposed method significantly outperforms state-of-the-art approaches. Codes data are available at https://github.com/MengHao666/HDR .