Deocclusion and integration of advantages for a better hand pose

Xuefeng Li, Xiangbo Lin

Abstract: Highlights•Generating bare hand from occluded hand image in image level.•A self-supervised deocclusion model without ground truth bare hand image.•3D hand–object interaction pose estimation without knowing object’s shape/pose.•Integrate efficient features from occluded and bare hand image by random fusion.•Achieve high hand pose estimation accuracy on HO3D and DexYCB datasets.