Research map / 3D vision / Human pose estimation
Zolly: Zoom Focal Length Correctly for Perspective-Distorted Human Mesh Reconstruction
Wenjia Wang, Yongtao Ge, Haiyi Mei, Zhongang Cai, Qingping Sun, Yanjun Wang, Chunhua Shen, Lei Yang, Taku Komura
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 distortion-feature-based translation and mesh model (Zolly) instead of fixed-focal HMR | HMR (R50) · PDHuman tau=3.0 | MPJPE (mm): 91.5 -> 76.4 | Improved |
| estimate focal length with SPEC network instead of fixed focal length | HMR · PDHuman tau=3.0 | MPJPE (mm): 91.5 -> 94.9 | Hurt |
| train HMR with ground-truth focal length (HMR-f) | HMR (R50) · PDHuman tau=3.0 | MPJPE (mm): 91.5 -> 90.2 | No clear effect |
| add distortion feature concatenated per vertex (ablation) | Zolly · PDHuman tau=3.0 | MPJPE (mm): 83.3 -> 76.4 | Improved |
| warp distortion image into UV space (ablation) | Zolly · PDHuman tau=3.0 | MPJPE (mm): 86.8 -> 83.3 | Improved |
| add perspective loss to weak-perspective loss (ablation) | Zolly · PDHuman tau=3.0 | mIoU: 47.3 -> 52.5 | Improved |
| add perspective loss to weak-perspective loss (ablation) | Zolly · PDHuman tau=3.0 | MPJPE (mm): 80.0 -> 79.1 | Improved |
| add per-joint distortion weight to weak-perspective loss (ablation) | Zolly · PDHuman tau=3.0 | MPJPE (mm): 79.1 -> 76.4 | Improved |
| add perspective-distorted training datasets (PDHuman, HuMMan) (ablation) | Zolly-H48 · 3DPW | PA-MPJPE (mm): 48.3 -> 47.9 | Improved |
| use ground-truth focal length when fine-tuning on 3DPW (ablation) | Zolly-H48 · 3DPW | PA-MPJPE (mm): 40.9 -> 39.8 | Improved |
| Zolly-H48 versus HMR-R50 on more distorted 3DPW subsets | HMR-R50 · 3DPW test protocols p1 to p3 | PA-MPJPE improvement (mm): +10.4 (p1) -> +13.6 (p3) | Improved |
| Zolly-H48 versus CLIFF on Human3.6M MPJPE | CLIFF (HRNet-w48) · Human3.6M | MPJPE (mm): 47.1 vs 49.4 | Hurt |
Closest papers by meaning
- BLADE: Single-view Body Mesh Estimation through Accurate Depth Estimation (CVPR 2025 · similarity 0.89)
- Recovering 3D Human Mesh From Monocular Images: A Survey (TPAMI 2023 · similarity 0.86)
- SPEC: Seeing People in the Wild With an Estimated Camera (ICCV 2021 · similarity 0.85)
- PyMAF-X: Towards Well-Aligned Full-Body Model Regression From Monocular Images (TPAMI 2023 · similarity 0.85)
- Reconstructing People, Places, and Cameras (CVPR 2025 · similarity 0.85)
- ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos (CVPR 2025 · similarity 0.84)
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.
