Research map / 3D vision / Human pose estimation
Contact and Human Dynamics from Monocular Video
Davis Rempe, Leonidas J. Guibas, Aaron Hertzmann, Bryan Russell, Ruben Villegas, Jimei Yang
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 |
|---|---|---|---|
| use network-predicted contacts instead of a velocity heuristic (ablation) | full optimization pipeline · synthetic test set | % of videos converged: 69.2% -> 94.9% | Improved |
| replace 2D velocity heuristic with learned MLP on lower-body joints | contact estimation · synthetic test set | accuracy: 0.853 -> 0.941 | Improved |
| replace 3D velocity heuristic with learned MLP on lower-body joints | contact estimation · real videos | accuracy: 0.875 -> 0.935 | Improved |
| add lower-body joints (up to hips) instead of upper-body joints down to hips (ablation) | contact MLP · real videos | accuracy: 0.692 -> 0.935 | Improved |
| use upper-body joints down to knees instead of lower-body joints (ablation) | contact MLP · real videos | accuracy: 0.935 -> 0.865 | Hurt |
| increase input window from 3 to 9 frames (prediction window 5) (ablation) | contact MLP · real videos | accuracy: 0.919 -> 0.935 | Improved |
| vary prediction window size (ablation) | contact MLP · contact estimation | accuracy: 9 input: 0.936 (3), 0.941 (5), 0.942 (7), 0.946 (9) synthetic | No clear effect |
| add physics-based optimization after kinematic initialization | Kinematics (ours) · synthetic test set | Max GRF: 1237.5% -> 338.6% | Improved |
| add physics-based optimization after kinematic initialization | Kinematics (ours) · synthetic test set | Ballistic GRF: 255.2% -> 0.0% | Improved |
| add physics-based optimization after kinematic initialization | Kinematics (ours) · synthetic test set | Floating: 2.3% -> 8.2% | Hurt |
| add physics-based optimization after kinematic initialization | Kinematics (ours) · synthetic test set | Skate: 1.6% -> 3.6% | Hurt |
| add physics-based optimization after kinematic initialization | Kinematics (ours) · synthetic test set | Penetration: 2.8% -> 0.3% | Improved |
Closest papers by meaning
- Trajectory Optimization for Physics-Based Reconstruction of 3D Human Pose From Monocular Video (CVPR 2022 · similarity 0.94)
- D&D: Learning Human Dynamics from Dynamic Camera (ECCV 2022 · similarity 0.92)
- Optimal-state Dynamics Estimation for Physics-based Human Motion Capture from Videos (NeurIPS 2024 · similarity 0.90)
- PhysPT: Physics-aware Pretrained Transformer for Estimating Human Dynamics from Monocular Videos (CVPR 2024 · similarity 0.89)
- Lifting Motion to the 3D World via 2D Diffusion (CVPR 2025 · similarity 0.88)
- Capturing Closely Interacted Two-Person Motions with Reaction Priors (CVPR 2024 · similarity 0.88)
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.
