English

Realistic Full-Body Tracking from Sparse Observations via Joint-Level Modeling

Computer Vision and Pattern Recognition 2023-08-21 v1

Abstract

To bridge the physical and virtual worlds for rapidly developed VR/AR applications, the ability to realistically drive 3D full-body avatars is of great significance. Although real-time body tracking with only the head-mounted displays (HMDs) and hand controllers is heavily under-constrained, a carefully designed end-to-end neural network is of great potential to solve the problem by learning from large-scale motion data. To this end, we propose a two-stage framework that can obtain accurate and smooth full-body motions with the three tracking signals of head and hands only. Our framework explicitly models the joint-level features in the first stage and utilizes them as spatiotemporal tokens for alternating spatial and temporal transformer blocks to capture joint-level correlations in the second stage. Furthermore, we design a set of loss terms to constrain the task of a high degree of freedom, such that we can exploit the potential of our joint-level modeling. With extensive experiments on the AMASS motion dataset and real-captured data, we validate the effectiveness of our designs and show our proposed method can achieve more accurate and smooth motion compared to existing approaches.

Keywords

Cite

@article{arxiv.2308.08855,
  title  = {Realistic Full-Body Tracking from Sparse Observations via Joint-Level Modeling},
  author = {Xiaozheng Zheng and Zhuo Su and Chao Wen and Zhou Xue and Xiaojie Jin},
  journal= {arXiv preprint arXiv:2308.08855},
  year   = {2023}
}

Comments

Accepted to ICCV 2023. Project page: https://zxz267.github.io/AvatarJLM

R2 v1 2026-06-28T11:57:46.542Z