The authors propose a generative framework that treats video motion capture as a conditional generation task.
Flow matching plus physics yields more trackable 3D human motion from video
A new framework, FlowHMR, uses a flow matching model pretrained on synthetic data and post-trained with reinforcement learning to generate motions that a physics-based controller can track, achieving 82.5% tracking success.
Chinese Tech
Zhanke Wang · Chengfeng Zhao · Qing Shuai · Jingzhong Lin · Heng Li · Zeyu Ling · +5 more
Peking University · The Hong Kong University of Science and Technology · Tencent · East China Normal University · Sun Yat-sen University
Research Digest··3 min read
Wang et al.
Why this paper
From Tencent and 5 others
In one line
FlowHMR uses flow matching with reinforcement learning to recover physically plausible global 3D human motion from monocular video, beating prior methods in fidelity and tracking.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§