Research map / 3D vision / Human pose estimation
TORE: Token Reduction for Efficient Human Mesh Recovery with Transformer
Zhiyang Dou, Qingxuan Wu, Cheng Lin, Zeyu Cao, Qiangqiang Wu, Weilin Wan, Taku Komura, Wenping Wang
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 Geometry Token Reduction (GTR) to METRO with ResNet-50 | METRO-R50 · Human3.6M | GFLOPs: 31.6 -> 5.4 | Improved |
| add GTR to METRO with ResNet-50 | METRO-R50 · Human3.6M | PAMPJPE (mm): 40.6 -> 42.0 | Hurt |
| add GTR to FastMETRO ResNet-50 | FastMETRO-R50 · Human3.6M | throughput (im/s): 576.0 -> 805.3 | Improved |
| add GTR to FastMETRO(S) / EfficientNet-b0 | FastMETRO(S)-R50, FastMETRO-Eb0 · Human3.6M | PAMPJPE (mm): 39.4 -> 38.6 and 45.8 -> 44.2 | Improved |
| add Image Token Pruning (ITP@20%) to FastMETRO-H64+GTR (ablation) | FastMETRO-H64+GTR · 3DPW | PAMPJPE (mm): 46.7 -> 44.4 | Improved |
| add ITP@20% to FastMETRO-R50+GTR (ablation) | FastMETRO-R50+GTR · Human3.6M | PAMPJPE (mm): 38.6 -> 40.5 | Hurt |
| add ITP to FastMETRO-Eb0+GTR (ablation) | FastMETRO-Eb0+GTR · Human3.6M | PAMPJPE (mm): 44.2 -> 43.9 | Improved |
| replace TokenLearner with ITP (ablation) | FastMETRO-R50+GTR · Human3.6M | PAMPJPE (mm): 45.4 -> 40.5 | Improved |
| replace MLP or GCN with Neural Shape Regressor (ablation) | GTR · 3DPW MPVE | MPVE (mm): 99.0 -> 95.9 | Improved |
| add token reduction supervision (projected mesh mask) (ablation) | ITP · Human3.6M | PAMPJPE (mm): 41.6 -> 40.6 | Improved |
| prune image tokens to one token (ablation) | ITP · Human3.6M | PAMPJPE (mm): 38.6 -> 43.3 | Hurt |
| apply GTR+ITP to hand mesh recovery | FastMETRO-H64 · FreiHAND | PAMPJPE (mm): 6.5 -> 6.7 | No clear effect |
Closest papers by meaning
- TokenHMR: Advancing Human Mesh Recovery with a Tokenized Pose Representation (CVPR 2024 · similarity 0.85)
- A Simple Baseline for Efficient Hand Mesh Reconstruction (CVPR 2024 · similarity 0.84)
- MEGA: Masked Generative Autoencoder for Human Mesh Recovery (CVPR 2025 · similarity 0.84)
- GenHMR: Generative Human Mesh Recovery (AAAI 2025 · similarity 0.84)
- How Many Tokens Do 3D Point Cloud Transformer Architectures Really Need? (NeurIPS 2025 · similarity 0.84)
- Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers (NeurIPS 2026 · 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.
