Research map / 3D vision / Human pose estimation
SMPLest-X: Ultimate Scaling for Expressive Human Pose and Shape Estimation
Wanqi Yin, Zhongang Cai, Ruisi Wang, Ailing Zeng, Wei Chen, Qingping Sun, Haiyi Mei, Yanjun Wang, Hui En Pang, Mingyuan Zhang, Lei Zhang, Chen Change Loy and 3 more
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 |
|---|---|---|---|
| scale training data from 5 to 40 datasets (ablation) | SMPLer-X-L · Whole-body MPE (5 benchmarks) | MPE: 97.0 -> 63.7 | Improved |
| scale training data from 5 to 40 datasets (ablation) | SMPLest-X-H · Whole-body MPE (5 benchmarks) | MPE: 87.5 -> 58.9 | Improved |
| increase instances from 7.5M to 10M within same 40 datasets (ablation) | SMPLest-X-H · Whole-body MPE (5 benchmarks) | MPE: 58.9 -> 59.9 | No clear effect |
| scale backbone from ViT-S to ViT-H (ablation) | SMPLer-X-40+ · Whole-body MPE (5 benchmarks) | MPE: 77.3 -> 59.9 | Improved |
| scale backbone from ViT-L to ViT-H (ablation) | SMPLer-X-40 · Whole-body MPE (5 benchmarks) | MPE: 63.7 -> 61.9 | Improved |
| replace component guiding module with transformer decoder and extra task tokens (SMPLest-X) | SMPLer-X-H40+ · Hand MPE (6 benchmarks) | hand-MPE: 35.6 -> 31.6 | Improved |
| replace component guiding module with transformer decoder and extra task tokens (SMPLest-X) | SMPLer-X-H40+ · Hand PA-MPE (6 benchmarks) | hand-PA-MPE: 11.1 -> 9.3 | Improved |
| select top-ranked datasets instead of bottom-ranked, 5 datasets (ablation) | SMPLer-X-B · Whole-body MPE (5 benchmarks) | MPE: 115.61 -> 103.47 | Improved |
| select top-ranked datasets instead of bottom-ranked, 10 datasets (ablation) | SMPLer-X-B · Whole-body MPE (5 benchmarks) | MPE: 115.10 -> 89.20 | Improved |
| weighted sampling instead of balanced sampling (ablation) | SMPLer-X-H32 · Whole-body MPE (5 benchmarks) | MPE: 63.08 -> 62.12 | No clear effect |
| concatenated sampling instead of balanced sampling (ablation) | SMPLer-X-H32 · Whole-body MPE (5 benchmarks) | MPE: 63.08 -> 63.32 | No clear effect |
| scale training data to 4.5M instances with CNN backbone (balanced) (ablation) | Hand4Whole · Whole-body MPE (5 benchmarks) | MPE: 116.59 -> 98.33 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- AiOS — AiOS: All-in-One-Stage Expressive Human Pose and Shape Estimation (CVPR 2024) · better on 4 of 4 shared results
- HybrIK-X — HybrIK-X: Hybrid Analytical-Neural Inverse Kinematics for Whole-Body Mesh Recovery (TPAMI 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SMPLer-X: Scaling Up Expressive Human Pose and Shape Estimation (NeurIPS 2023 · similarity 0.97)
- Expressive Body Capture: 3D Hands, Face, and Body From a Single Image (CVPR 2019 · similarity 0.89)
- Monocular Expressive Body Regression through Body-Driven Attention (ECCV 2020 · similarity 0.89)
- PoseBH: Prototypical Multi-Dataset Training Beyond Human Pose Estimation (CVPR 2025 · similarity 0.87)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.86)
- AiOS: All-in-One-Stage Expressive Human Pose and Shape Estimation (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.
