Research map / 3D vision / Human pose estimation
Bias-Compensated Integral Regression for Human Pose Estimation
Kerui Gu, Linlin Yang, Michael Bi Mi, Angela Yao
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 bias compensation (de-bias) (ablation) | integral regression (IPR) · COCO val (SBL-ResNet50) | AP: 68.2 -> 71.0 | Improved |
| add Gaussian prior loss L_de (first 120 epochs) (ablation) | integral regression (IPR) · COCO val (SBL-ResNet50) | AP: 68.2 -> 70.4 | Improved |
| add Laplacian heatmap shrinkage regularizer L_reg (ablation) | integral regression (IPR) · COCO val (SBL-ResNet50) | AP: 68.2 -> 69.5 | Improved |
| apply L_de for all epochs instead of 120 (ablation) | integral regression + L_de · COCO val (SBL-ResNet50) | AP: 70.4 -> 70.2 | Hurt |
| combine L_de and de-bias (BCIR) (ablation) | integral regression (IPR) · COCO val (SBL-ResNet50) | AP: 68.2 -> 71.2 | Improved |
| set beta to 1 instead of 10 (ablation) | BCIR · COCO val (SBL-ResNet50) | AP: 71.2 -> 70.7 | Hurt |
| set beta to 20 instead of 10 (ablation) | BCIR · COCO val (SBL-ResNet50) | AP: 71.2 -> 70.3 | Hurt |
| add BCIR to integral regression | SBL-ResNet50 · COCO val | AP: 68.2 -> 71.2 | Improved |
| add BCIR to integral regression | HRNet-W32 · COCO val | AP: 72.9 -> 75.8 | Improved |
| add BCIR to integral regression | SBL-ResNet50 · MPII val | PCKh@0.5: 86.2 -> 87.9 | Improved |
| add BCIR to 2.5D regression | 2.5D regression · RHD test | 2D AUC: 84.4 -> 85.8 | Improved |
| use integral regression instead of detection on hard samples (1-5 joints, 32-64px) | SBL · COCO val | EPE: 9.23 -> 8.10 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Poseur — Poseur: Direct Human Pose Regression with Transformers (ECCV 2022) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Dive Deeper Into Integral Pose Regression (ICLR 2022 · similarity 0.94)
- Removing the Bias of Integral Pose Regression (ICCV 2021 · similarity 0.93)
- Integral Human Pose Regression (ECCV 2018 · similarity 0.89)
- Heatmap Distribution Matching for Human Pose Estimation (NeurIPS 2022 · similarity 0.87)
- Continuous Heatmap Regression for Pose Estimation via Implicit Neural Representation (NeurIPS 2024 · similarity 0.86)
- Video-Based Human Pose Regression via Decoupled Space-Time Aggregation (CVPR 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.
