UPSR: a Unified Proxy Skeleton Retargeting Method for Heterogeneous Avatar Animation

Wenfeng Song, Xinyu Zhang, Yang Gao, Yifan Luo, Haoxiang Wang, Xianfei Wang, Xia Hou

Abstract: Pose-driven avatar animation is widely applied in VR fields. The flexible creation of animations is a significant yet challenging task due to the heterogeneous topologies and shapes of avatars. To allevi-ate this, we propose a unified proxy skeleton for retargeting (UPSR) to achieve consistent motions of the virtual avatar. Particularly, the heterogeneous topologies are converted into a unified skeleton topol-ogy of the avatars by using a learned nonlinear mapping function. Furthermore, we propose to retarget the skeletons with different body shapes into 3D virtual avatars. Our UPSR can produce avatar animations with a higher level of authenticity without the depen-dency on high-cost motion capture devices or the restrictions of its topology. Additionally, we could drive detailed motions based on multiple sources of motion datasets, including monocular videos, motion capture devices, and standard motion files, with the precise motion capture of eyes, lips, hands, and bodies.