Research map / Video / Video generation
A Plug-and-Play Physical Motion Restoration Approach for In-the-Wild High-Difficulty Motions
Youliang Zhang, Ronghui Li, Yachao Zhang, Liang Pan, Jingbo Wang, Yebin Liu, Xiu Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add MCM and physics fine-tuning with mask guidance (ablation) | PTM pipeline · High-difficulty in-the-wild dataset | success rate: 65% -> 87% | Improved |
| use human masks instead of 2D keypoints as guidance (ablation) | MCM and PTM · High-difficulty in-the-wild dataset | OKS, MPS, SR | Improved |
| add motion prior initialization (ablation) | PTM adaptation · High-difficulty in-the-wild dataset | success rate: 77% -> 87% | Improved |
| add motion prior and residual action (ablation) | PTM adaptation · High-difficulty in-the-wild dataset | average adaptation steps: 4.5k -> 2.5k | Improved |
| replace PHC+ with pretrained PTM before adaptation (ablation) | PHC+ · High-difficulty in-the-wild dataset | success rate: 21% -> 37% | Improved |
| add the plug-and-play restoration to TRAM | TRAM · High-difficulty in-the-wild dataset | ground penetration: 19.988 -> 0.595 | Improved |
| add the plug-and-play restoration to GVHMR | GVHMR · High-difficulty in-the-wild dataset | OKS: 0.837 -> 0.865 | Improved |
| add PhysPT post-processing to GVHMR | GVHMR · High-difficulty in-the-wild dataset | OKS: 0.837 -> 0.806 | Hurt |
| add restoration to T2M-GPT | T2M-GPT · HumanML3D | FID: 0.116 -> 0.119 | No clear effect |
| add restoration to Momask | Momask · HumanML3D | FID: 0.045 -> 0.062 | Hurt |
| add restoration to Momask | Momask · HumanML3D | ground penetration: 20.672 -> 0.233 | Improved |
| use PTM instead of PHC+ for imitation | PHC+ · AIST++ and Kungfu | success rate: 72.74% -> 96.87% | Improved |
Closest papers by meaning
- Natural Human Motion Recovery by Aligning High-Order Temporal Dynamics from Monocular Videos (CVPR 2026 · similarity 0.88)
- HumanMM: Global Human Motion Recovery from Multi-shot Videos (CVPR 2025 · similarity 0.88)
- Physics-Based Human Motion Estimation and Synthesis From Videos (ICCV 2021 · similarity 0.88)
- Mimic-X: A Large-Scale Motion Dataset via Fast Physics-Based Controller Adaptation (AAAI 2026 · similarity 0.88)
- SurMo: Surface-based 4D Motion Modeling for Dynamic Human Rendering (CVPR 2024 · similarity 0.86)
- Optimal-state Dynamics Estimation for Physics-based Human Motion Capture from Videos (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
