Research map / 3D vision / 3D generation
Monocular Models are Strong Learners for Multi-View Human Mesh Recovery
Haoyu Xie, Shengkai Xu, Cheng Guo, Muhammad Saleem, Wenhan Wu, Chen Chen, Ahmed Helmy, Pu Wang, Hongfei Xue
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 test-time adaptation to virtual-view initialization (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 44.2 -> 26.9 | Improved |
| Add virtual view to TTA (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 27.8 -> 26.9 | Improved |
| Add virtual view to TTA (ablation) | TokenHMR backbone · Human3.6M | MPJPE: 43.9 -> 32.6 | Improved |
| Use weighted init instead of T-pose init (ablation) | TokenHMR backbone · Human3.6M | MPJPE: 33.8 -> 32.6 | Improved |
| Add multi-view consistency loss to kp+ana (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 36.8 -> 27.3 | Improved |
| Add regularization loss (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 27.3 -> 26.9 | Improved |
| Optimize learned SMPL tokens instead of explicit SMPL parameters (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 58.2 -> 26.9 | Improved |
| Optimize learned tokens instead of decoder weights (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 30.8 -> 26.9 | Improved |
| Also optimize orientation and shape in virtual view (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 31.6 -> 26.9 | Improved |
| Increase number of views from 2 to 4 (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 31.2 -> 26.9 | Improved |
| Increase learning rate from 6e-2 to 1e-1 (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 26.9 -> 29.5 | Hurt |
| Increase TTA steps from 200 to 250 (ablation) | CameraHMR backbone · Human3.6M | MPJPE: 26.9 -> 27.4 | No clear effect |
Closest papers by meaning
- Progressive Multi-View Human Mesh Recovery with Self-Supervision (AAAI 2023 · similarity 0.92)
- FactorizedHMR: A Hybrid Framework for Video Human Mesh Recovery (NeurIPS 2026 · similarity 0.90)
- ADHMR: Aligning Diffusion-based Human Mesh Recovery via Direct Preference Optimization (ICML 2025 · similarity 0.90)
- GenHMR: Generative Human Mesh Recovery (AAAI 2025 · similarity 0.89)
- Incorporating Test-Time Optimization into Training with Dual Networks for Human Mesh Recovery (NeurIPS 2024 · similarity 0.89)
- BLADE: Single-view Body Mesh Estimation through Accurate Depth Estimation (CVPR 2025 · similarity 0.89)
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.
