Date: June 2023.
Source: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Proceedings Page(s): 4670-4680. DOI Bookmark: 10.1109/CVPR52729.2023.00453
Research Summary: Handy addresses the lack of diversity and appearance detail in existing 3D hand models, most of which rely on the MANO model, which is trained on only 31 adult subjects. Handy is a large-scale parametric model of hand shape and texture trained from real-world data of over 1,200 subjects spanning ages 1 to 81, with a StyleGAN-based architecture trained to capture high-frequency texture detail such as wrinkles, veins, and nail polish. The model outperforms existing methods on shape reconstruction, texture fidelity, and generalization to children’s hands, a population absent from prior hand models.
3dMD’s Role: Handy’s dataset was captured on a 3dMDhand5 system, an active stereo photogrammetry, software-driven optics-based system, during the same 2019/2020 4D Body and Hand Capture Exercise at the Science Museum, London that produced the 3DH hand dataset. The 1,208 subjects span ages 1 to 81 and the scans provided both the real-world 3D shape and texture data behind the Handy model.
Article: Handy: Towards a high fidelity 3D hand shape and appearance model.
Authors: Rolandos Alexandros Potamias, Stylianos Ploumpis, Stylianos Moschoglou, Vasileios Triantafyllou, Stefanos Zafeiriou, Imperial College London.