Research map / 3D vision / Human pose estimation
Heatmap Regression via Randomized Rounding
Baosheng Yu, 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 2 activation points instead of 1 (ablation) | H3R · WFLW 256x256 binary heatmap | NME (%): 4.005 -> 3.881 | Improved |
| use 9 activation points instead of 1 (ablation) | H3R · WFLW 256x256 binary heatmap | NME (%): 4.005 -> 3.807 | Improved |
| use 2 activation points instead of 1 at low resolution (ablation) | H3R · WFLW 128x128 | NME (%): 4.637 -> 4.164 | Improved |
| use more than 7 activation points at 128x128 resolution (ablation) | H3R · WFLW 128x128 | NME (%): 3.988 -> 4.002 | Hurt |
| add more activation points with sigma=0 Gaussian (ablation) | H3R · WFLW sigma=0.0 | NME (%): 4.065 -> 6.802 | Hurt |
| use Gaussian heatmap sigma=1.0 with k=11 instead of binary (ablation) | H3R · WFLW | NME (%): 3.988 -> 3.855 | Improved |
| train and test with box policy P1 instead of P2 (ablation) | H3R · WFLW test | NME (%): 3.95 -> 3.81 | Improved |
| mismatch box policy at test time (P1 train, P2 test) (ablation) | H3R · WFLW test | NME (%): 3.81 -> 5.34 | Hurt |
| add H3R instead of vanilla k=1 (ablation) | HRNet-W32 · MPII PCKh@0.1 | PCKh@0.1: 32.8 -> 39.3 | Improved |
| add H3R instead of vanilla k=1 (ablation) | HRNet-W32 · MPII PCKh@0.5 mean | PCKh@0.5: 90.1 -> 90.3 | No clear effect |
| use H3R rather than quarter-shift compensation (ablation) | HRNet-W32 · COCO 256x192 | AP: 0.744 -> 0.750 | Improved |
| add H3R instead of vanilla k=1 (ablation) | HRNet-W32 · COCO 192x128 | AP: 0.674 -> 0.720 | Improved |
Closest papers by meaning
- Heatmap Regression without Soft-Argmax for Facial Landmark Detection (ICCV 2025 · similarity 0.87)
- Adaptive Wing Loss for Robust Face Alignment via Heatmap Regression (ICCV 2019 · similarity 0.87)
- Heatmap Distribution Matching for Human Pose Estimation (NeurIPS 2022 · similarity 0.86)
- POPoS: Improving Efficient and Robust Facial Landmark Detection with Parallel Optimal Position Search (AAAI 2025 · similarity 0.84)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.84)
- Continuous Heatmap Regression for Pose Estimation via Implicit Neural Representation (NeurIPS 2024 · 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.
