Research map / 3D vision / Human pose estimation
TransPose: Keypoint Localization via Transformer
Sen Yang, Zhibin Quan, Mu Nie, Wankou Yang
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 Transformer encoder on shallow ResNet backbone (TransPose-R-A4 vs SimpleBaseline-Res152) | ResNet-S · COCO val2017 | AP: 72.0 -> 72.6 | Improved |
| add Transformer encoder on shallow HRNet backbone (TransPose-H-A6 vs HRNet-W48) | HRNet-S · COCO val2017 | AP: 75.1 -> 75.8 | Improved |
| add position embedding (2D sine) to TransPose-R-A3 (ablation) | TransPose-R-A3 · COCO val2017 | AP: 70.4 -> 71.7 | Improved |
| use learnable instead of no position embedding (ablation) | TransPose-R-A3 · COCO val2017 | AP: 70.4 -> 70.9 | Improved |
| use fixed 2D sine instead of learnable position embedding (ablation) | TransPose-R-A3 · COCO val2017 | AP: 70.9 -> 71.7 | Improved |
| add encoder layers from 2 to 4 in TransPose-R (ablation) | TransPose-R · COCO val2017 | AP: 69.6 -> 72.6 | Improved |
| add encoder layers from 4 to 6 in TransPose-R (ablation) | TransPose-R · COCO val2017 | AP: 72.6 -> 72.2 | Hurt |
| add encoder layers from 4 to 6 in TransPose-H (ablation) | TransPose-H · COCO val2017 | AP: 75.3 -> 75.8 | Improved |
| increase embedding dimension d from 64 to 192 in TransPose-H (ablation) | TransPose-H · COCO val2017 | AP: 75.1 -> 75.4 | Improved |
| COCO pre-training then fine-tuning on MPII instead of full-training (ablation) | TransPose-R-A4 · MPII val | Mean@0.5: 89.3 -> 92.0 | Improved |
| COCO pre-training then fine-tuning on MPII instead of full-training (ablation) | TransPose-H-A6 · MPII val | Mean@0.5: 90.3 -> 92.3 | Improved |
| COCO pre-training then fine-tuning on MPII instead of full-training (ablation) | DARK-HRNet · MPII val | Mean@0.5: 90.6 -> 92.0 | Improved |
Closest papers by meaning
- ProbPose: A Probabilistic Approach to 2D Human Pose Estimation (CVPR 2025 · similarity 0.89)
- Few-Shot Geometry-Aware Keypoint Localization (CVPR 2023 · similarity 0.88)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.87)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.86)
- HyperPose: Hypernetwork-Infused Camera Pose Localization and an Extended Cambridge Landmarks Dataset (CVPR 2025 · similarity 0.86)
- SDPose: Tokenized Pose Estimation via Circulation-Guide Self-Distillation (CVPR 2024 · 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.
