Research map / 3D vision / Human pose estimation
VLM-Guided Group Preference Alignment for Diffusion-based Human Mesh Recovery
Wenhao Shen, Hao Wang, Wanqi Yin, Fayao Liu, Xulei Yang, Chao Liang, Zhongang Cai, Guosheng Lin
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 |
|---|---|---|---|
| replace HMR-Scorer with the VLM critique agent to build the preference dataset (ablation) | group preference alignment · 3DPW (InstaVariety, M=100) | MPJPE: 54.9 -> 49.9 | Improved |
| use group-wise advantage alignment instead of pairwise DPO (ablation) | diffusion HMR with critique agent · 3DPW (InstaVariety, M=100) | MPJPE: 53.1 -> 49.9 | Improved |
| use preference alignment instead of supervised finetuning on pseudo-labels (ablation) | base diffusion HMR model · 3DPW (InstaVariety, M=100) | MPJPE: 61.3 -> 49.9 | Improved |
| add rule memory to the critique agent (ablation) | critique agent · GTA-Human II group-wise (PVE) | SRCC: 0.597 -> 0.605 | Improved |
| add prototype memory to the critique agent (ablation) | critique agent · GTA-Human II group-wise (PVE) | SRCC: 0.590 -> 0.605 | Improved |
| add self-reflection rule mining to the critique agent (ablation) | critique agent · GTA-Human II group-wise (PVE) | SRCC: 0.534 -> 0.605 | Improved |
| add self-reflection rule mining to the critique agent (ablation) | critique agent · DNA-Rendering group-wise (PVE) | SRCC: 0.568 -> 0.633 | Improved |
| add self-reflection to the critique agent (ablation) | critique agent · GTA-Human II point-wise (PVE) | PLCC: 0.610 -> 0.695 | Improved |
| add prototype memory to the critique agent (ablation) | critique agent · GTA-Human II point-wise (PVE) | PLCC: 0.645 -> 0.695 | Improved |
| add rule memory to the critique agent (ablation) | critique agent · GTA-Human II point-wise (PVE) | PLCC: 0.683 -> 0.695 | Improved |
| increase number of hypotheses M from 10 to 100 to 200 | Ours · 3DPW | MPJPE: 59.3 -> 52.5 -> 51.0 | Improved |
| finetune with group preference alignment instead of ADHMR DPO | diffusion HMR · 3DPW | MPJPE: 57.2 -> 52.5 | Improved |
Closest papers by meaning
- ADHMR: Aligning Diffusion-based Human Mesh Recovery via Direct Preference Optimization (ICML 2025 · similarity 0.93)
- GenHMR: Generative Human Mesh Recovery (AAAI 2025 · similarity 0.88)
- DPMesh: Exploiting Diffusion Prior for Occluded Human Mesh Recovery (CVPR 2024 · similarity 0.88)
- Score-Guided Diffusion for 3D Human Recovery (CVPR 2024 · similarity 0.88)
- Incorporating Test-Time Optimization into Training with Dual Networks for Human Mesh Recovery (NeurIPS 2024 · similarity 0.88)
- FactorizedHMR: A Hybrid Framework for Video Human Mesh Recovery (NeurIPS 2026 · similarity 0.87)
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.
