Research map / 3D vision / Human pose estimation
AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking in Real-Time
Hao-Shu Fang, Jiefeng Li, Hongyang Tang, Chao Xu, Haoyi Zhu, Yuliang Xiu, Yong–Lu Li, 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 |
|---|---|---|---|
| use two-step heatmap normalization instead of one-step soft-max (ablation) | FastPose50 · Halpe-FullBody | mAP: 38.1 -> 44.1 | Improved |
| use two-step heatmap normalization instead of one-step soft-max (ablation) | FastPose50 · COCO val | mAP: 67.1 -> 69.5 | Improved |
| add symmetric integral keypoints regression (SIKR) (ablation) | FastPose50 · Halpe-FullBody | mAP: 42.3 -> 44.1 | Improved |
| add symmetric integral keypoints regression (SIKR) (ablation) | FastPose50 · COCO val | mAP: 64.6 -> 69.5 | Improved |
| add parametric pose NMS (ablation) | FastPose50 · Halpe-FullBody | mAP: 43.7 -> 44.1 | Improved |
| add parametric pose NMS (ablation) | FastPose50 · COCO val | mAP: 68.2 -> 69.5 | Improved |
| add part-guided proposal generator (PGPG) (ablation) | FastPose50-dcn · Halpe-FullBody | mAP: 47.1 -> 48.4 | Improved |
| use symmetric integral regression instead of heatmap | FastPose50 · Halpe-FullBody (full-body) | AP: 0.417 -> 0.441 | Improved |
| use symmetric integral regression instead of heatmap | FastPose · COCO test-dev (body) | AP: 0.718 -> 0.649 | Hurt |
| add deformable convolution to the backbone | FastPose50-si · Halpe-FullBody (full-body) | AP: 0.441 -> 0.462 | Improved |
| add multi-domain knowledge distillation and PGPG training (ablation) | FastPose50-dcn-si · Halpe-FullBody (full-body) | AP: 0.462 -> 0.484 | Improved |
| use ResNet152 instead of ResNet50 backbone | FastPose-si · Halpe-FullBody (full-body) | AP: 0.441 -> 0.451 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OpenPose — OpenPose: Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields (TPAMI 2019) · better on 1 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
- PoseTrack: Joint Multi-Person Pose Estimation and Tracking (CVPR 2017 · similarity 0.90)
- Single-Network Whole-Body Pose Estimation (ICCV 2019 · similarity 0.90)
- PoseTrack: A Benchmark for Human Pose Estimation and Tracking (CVPR 2018 · similarity 0.90)
- Detect-and-Track: Efficient Pose Estimation in Videos (CVPR 2018 · similarity 0.90)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.89)
- AiOS: All-in-One-Stage Expressive Human Pose and Shape Estimation (CVPR 2024 · similarity 0.88)
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.
