Research map / Label-efficient and robust learning / Knowledge distillation
DistilPose: Tokenized Pose Regression With Heatmap Distillation
Suhang Ye, Yingyi Zhang, Jie Hu, Liujuan Cao, Shengchuan Zhang, Lei Shen, Jun Wang, Shouhong Ding, Rongrong Ji
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 TDE using keypoint- and visual-token losses (ablation) | DistilPose-S · MSCOCO val | AP: +11.5% | Improved |
| add simulated heatmaps using confidence and heatmap losses (ablation) | DistilPose-S · MSCOCO val | AP: +8.1% | Improved |
| add confidence loss L_CS alone (ablation) | DistilPose-S · MSCOCO val | AP: +7.2% | Improved |
| add heatmap-matching loss L_SH alone (ablation) | DistilPose-S · MSCOCO val | AP: +0.4% | Improved |
| combine confidence and simulated-heatmap losses (ablation) | DistilPose-S · MSCOCO val | AP: +8.1% | Improved |
| add keypoint-token distillation L_KT alone (ablation) | DistilPose-S · MSCOCO val | AP: +11.1% | Improved |
| add visual-token distillation L_VT alone (ablation) | DistilPose-S · MSCOCO val | AP: +5.7% | Improved |
| combine keypoint- and visual-token distillation (ablation) | DistilPose-S · MSCOCO val | AP: +11.5% | Improved |
| combine all four distillation losses (ablation) | DistilPose-S · MSCOCO val | AP: +15.6% | Improved |
| use a stronger HRNet-W48 teacher instead of no teacher (ablation) | stemnet student · MSCOCO val | AP: 56.0% -> 71.6% | Improved |
| use a stronger HRNet-W48 teacher instead of no teacher (ablation) | HRNet-W48-stage3 student · MSCOCO val | AP: 63.0% -> 74.4% | Improved |
| add simulated heatmaps (ablation) | Deeppose with ResNet-50 · MSCOCO val | mAP: + 6.9% | Improved |
Closest papers by meaning
- Online Knowledge Distillation for Efficient Pose Estimation (ICCV 2021 · similarity 0.89)
- Human Pose Regression With Residual Log-Likelihood Estimation (ICCV 2021 · similarity 0.89)
- SDPose: Tokenized Pose Estimation via Circulation-Guide Self-Distillation (CVPR 2024 · similarity 0.87)
- Continuous Heatmap Regression for Pose Estimation via Implicit Neural Representation (NeurIPS 2024 · similarity 0.86)
- SHaRPose: Sparse High-Resolution Representation for Human Pose Estimation (AAAI 2024 · similarity 0.85)
- MotionMap: Representing Multimodality in Human Pose Forecasting (CVPR 2025 · similarity 0.85)
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.
