Research map / 3D vision / Human pose estimation
Human Pose Regression With Residual Log-Likelihood Estimation
Jiefeng Li, Siyuan Bian, Ailing Zeng, Can Wang, Bo Pang, Wentao Liu, Cewu Lu
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 |
|---|---|---|---|
| replace l1 loss with RLE | Direct regression ResNet-50+FC · COCO val | AP: 58.1 -> 70.5 | Improved |
| add residual log-likelihood estimation on top of reparameterized direct likelihood estimation (ablation) | Regression with DLE · COCO val | AP: 62.7 -> 70.5 | Improved |
| use Laplace instead of Gaussian as initial density Q in RLE (ablation) | RLE · COCO val | AP: 70.0 -> 70.5 | Improved |
| add RLE-learned density to Laplace with learned sigma (ablation) | Laplace baseline · COCO val | AP: 67.4 -> 70.5 | Improved |
| learn sigma instead of constant variance (Laplace) (ablation) | Direct regression · COCO val | AP: 58.1 -> 67.4 | Improved |
| increase RealNVP width from 64 to 256 (3 layers) (ablation) | RLE · COCO val | AP: 70.5 -> 69.6 | No clear effect |
| add RLE to HRNet-W32 with soft-argmax | HRNet-W32 · COCO val | AP: 74.1 -> 74.3 | Improved |
| add RLE to Mask R-CNN | Mask R-CNN · COCO val | AP: 66.0 -> 66.7 | Improved |
| add RLE to PointSet Anchor | PointSet Anchor · COCO val | AP: 67.0 -> 67.4 | Improved |
| add auxiliary heatmap loss (ablation) | RLE and direct regression · COCO val | AP: 70.5 -> 70.4 | No clear effect |
| replace heatmap confidence with RLE sigma-based confidence | SimplePose · COCO val | Pearson correlation with OKS: 0.479 -> 0.553 | Improved |
| add RLE to single-stage 3D regression | Direct regression · Human3.6M | MPJPE (mm): 50.1 -> 48.6 | Improved |
Closest papers by meaning
- DistilPose: Tokenized Pose Regression With Heatmap Distillation (CVPR 2023 · similarity 0.89)
- Rethinking the Heatmap Regression for Bottom-Up Human Pose Estimation (CVPR 2021 · similarity 0.88)
- Human Pose Estimation via Convolutional Part Heatmap Regression (ECCV 2016 · similarity 0.88)
- Removing the Bias of Integral Pose Regression (ICCV 2021 · similarity 0.88)
- Continuous Heatmap Regression for Pose Estimation via Implicit Neural Representation (NeurIPS 2024 · similarity 0.87)
- Video-Based Human Pose Regression via Decoupled Space-Time Aggregation (CVPR 2024 · similarity 0.86)
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.
