Research map / 3D vision / Human pose estimation
ViTPose++: Vision Transformer for Generic Body Pose Estimation
Yufei Xu, Jing Zhang, Qiming Zhang, Dacheng Tao
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 simple decoder instead of classic decoder with ResNet-50 (ablation) | SimpleBaseline ResNet-50 · MS COCO val | AP: 71.8 -> 53.1 | Hurt |
| use simple decoder instead of classic decoder with ViT backbone (ablation) | ViTPose-B · MS COCO val | AP: 75.8 -> 75.5 | No clear effect |
| use minimal decoder (single linear layer) (ablation) | ViTPose-B · MS COCO val | AP: 75.8 -> 75.4 | No clear effect |
| replace random init with MAE pre-training (ablation) | ViTPose-B · MS COCO val | AP: 72.8 -> 75.8 | Improved |
| replace MAE with supervised DeiT or contrastive MoCov3 pre-training (ablation) | ViTPose-B · MS COCO val | AP: 75.8 -> 72.7 (DeiT), 72.1 (MoCov3) | Hurt |
| increase input resolution from 256x192 to 576x432 (ablation) | ViTPose-B · MS COCO val | AP: 75.8 -> 77.8 | Improved |
| 1/8 feature size with full attention instead of window attention without shift (ablation) | ViTPose-B · MS COCO val | AP: 66.4 -> 77.4 | Improved |
| add shift windows to window attention (ablation) | ViTPose-B 1/8 feature · MS COCO val | AP: 66.4 -> 76.4 | Improved |
| freeze FFN during fine-tuning (ablation) | ViTPose-B · MS COCO val | AP: 75.8 -> 72.8 | Hurt |
| freeze MHSA during fine-tuning (ablation) | ViTPose-B · MS COCO val | AP: 75.8 -> 75.1 | Hurt |
| add knowledge token distillation from ViTPose-L (ablation) | ViTPose-B · MS COCO val | AP: 75.8 -> 76.0 | Improved |
| add output (heatmap) distillation from ViTPose-L (ablation) | ViTPose-B · MS COCO val | AP: 75.8 -> 76.3 | Improved |
Closest papers by meaning
- ViTPose: Simple Vision Transformer Baselines for Human Pose Estimation (NeurIPS 2022 · similarity 0.96)
- Transformers With Joint Tokens and Local-Global Attention for Efficient Human Pose Estimation (TPAMI 2026 · similarity 0.91)
- ExtPose: Robust and Coherent Pose Estimation by Extending ViTs (ICML 2025 · similarity 0.88)
- SHaRPose: Sparse High-Resolution Representation for Human Pose Estimation (AAAI 2024 · similarity 0.87)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.86)
- SDPose: Tokenized Pose Estimation via Circulation-Guide Self-Distillation (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.
