InterHand2.6M: A Dataset and Baseline for 3D Interacting Hand Pose Estimation from a Single RGB Image

Analysis of hand-hand interactions is a crucial step towards better\nunderstanding human behavior. However, most researches in 3D hand pose\nestimation have focused on the isolated single hand case. Therefore, we firstly\npropose (1) a large-scale dataset, InterHand2.6M, and (2) a baseline network,\nInterNet, for 3D interacting hand pose estimation from a single RGB image. The\nproposed InterHand2.6M consists of \\textbf{2.6M labeled single and interacting\nhand frames} under various poses from multiple subjects. Our InterNet\nsimultaneously performs 3D single and interacting hand pose estimation. In our\nexperiments, we demonstrate big gains in 3D interacting hand pose estimation\naccuracy when leveraging the interacting hand data in InterHand2.6M. We also\nreport the accuracy of InterNet on InterHand2.6M, which serves as a strong\nbaseline for this new dataset. Finally, we show 3D interacting hand pose\nestimation results from general images. Our code and dataset are available at\nhttps://mks0601.github.io/InterHand2.6M/.\n