Research map / 3D vision / Human pose estimation
Learning Motion Priors for 4D Human Body Capture in 3D Scenes
Siwei Zhang, Yan Zhang, Federica Bogo, Marc Pollefeys, Siyu Tang
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 the learned motion smoothness prior instead of L2-A baseline | per-frame fitting · PROX (205 clips) | PSKL-J (P,A): 0.481 -> 0.272 | Improved |
| add the learned motion smoothness prior (vs original PROX) | per-frame fitting · PROX (205 clips) | PSKL-J (P,A): 1.464 -> 0.272 | Improved |
| add smoothness prior | per-frame fitting · PROX (205 clips) | 2DJE: 20.94 -> 20.64 | Improved |
| add smoothness prior vs DCT on 3DPW | per-frame fitting · 3DPW | PSKL-J (3D,A): 0.242 -> 0.173 | Improved |
| add the motion infilling stage (Stage 3) after smoothing (ablation) | Ours-S2 · PROX (205 clips) | 2DJE: 20.40 -> 20.23 | Improved |
| add the motion infilling stage (Stage 3) after smoothing (ablation) | Ours-S2 · PROX (205 clips) | NonColl: 0.977 -> 0.979 | Improved |
| use the stage order smoothing then infilling (vs swapped) (ablation) | Ours-S3 · PROX (205 clips) | 2DJE: 20.64 -> 20.23 | Improved |
| use the stage order smoothing then infilling (vs swapped) (ablation) | Ours-S3 · PROX (205 clips) | PSKL-J (P,A): 0.307 -> 0.256 | Improved |
| add per-instance self-supervised finetuning (ablation) | Ours-IP · AMASS (130 test sequences, lower body masked) | MPMPE: 0.023 -> 0.014 | Improved |
| add per-instance self-supervised finetuning (ablation) | Ours-IP · AMASS (130 test sequences, lower body masked) | Foot skating: 0.178 -> 0.182 | Hurt |
| add SMPL-X optimization after infilling (ablation) | Ours-IP · AMASS (130 test sequences, lower body masked) | MPMPE: 0.015 -> 0.016 | Mixed |
| use predicted contact labels instead of heuristic contact (ablation) | Ours-IP · AMASS (130 test sequences, lower body masked) | Foot skating: 0.265 -> 0.182 | Improved |
Closest papers by meaning
- SurMo: Surface-based 4D Motion Modeling for Dynamic Human Rendering (CVPR 2024 · similarity 0.88)
- Contact and Human Dynamics from Monocular Video (ECCV 2020 · similarity 0.88)
- Monocular Total Capture: Posing Face, Body, and Hands in the Wild (CVPR 2019 · similarity 0.88)
- UniCon3R: Unified Contact-aware 4D Human-Scene Reconstruction from Monocular Video (NeurIPS 2026 · similarity 0.87)
- Pose Priors from Language Models (CVPR 2025 · similarity 0.86)
- Capturing Closely Interacted Two-Person Motions with Reaction Priors (CVPR 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.
