Research map / 3D vision / Human pose estimation
Whole-Body Human Pose Estimation in the Wild
Sheng Jin, Lumin Xu, Jin Xu, Can Wang, Wentao Liu, Chen Qian, Wanli Ouyang, Ping Luo
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 |
|---|---|---|---|
| train on all 133 whole-body keypoints instead of 17 body keypoints only | HRNet · COCO-WholeBody body keypoints | mAP: -9.9% mAP | Hurt |
| train on all 133 whole-body keypoints instead of 17 body keypoints only | AE · COCO-WholeBody body keypoints | mAP: -17.7% mAP | Hurt |
| train on all 133 whole-body keypoints instead of 17 body keypoints only | PAF · COCO-WholeBody body keypoints | mAP: -14.3% mAP | Hurt |
| two-stage zoom-in decoupling body and face/hand estimation | ZoomNet · COCO-WholeBody body keypoints | mAP: -1.5% mAP | Mixed |
| replace one-stage HRNet with zoom-in ZoomNet (same HRNet-W32 backbone) | HRNet · COCO-WholeBody | whole-body mAP: 10.9% mAP and 13.8% mAR | Improved |
| use ground-truth face/hand boxes instead of predicted boxes (ablation) | ZoomNet · COCO-WholeBody | face/hand/whole-body mAP: 19.6%, 8.4% and 23.6% | Improved |
| add joint training of BodyNet and Face/HandHead instead of reusing features (ablation) | ZoomNet · COCO-WholeBody | face AP: 0.609 -> 0.623 | Improved |
| add joint training of BodyNet and Face/HandHead instead of reusing features (ablation) | ZoomNet · COCO-WholeBody | hand AP: 0.393 -> 0.401 | Improved |
| use fully independent face/hand detectors instead of joint training (ablation) | ZoomNet · COCO-WholeBody | whole-body AP: 0.541 -> 0.543 | No clear effect |
| pretrain on WholeBody-Hand before fine-tuning on Panoptic (ablation) | HRNetV2-W18 · CMU Panoptic hand | EPE: 7.49 -> 7.00 | Improved |
| fine-tune on 300W after training on WholeBody-Face (ablation) | HRNetV2-W18 · 300W challenging | NME: 5.15 -> 4.73 | Improved |
| pretrain on Panoptic then fine-tune on WholeBody-Hand (ablation) | HRNetV2-W18 · WBH | NME: 6.66 -> 6.49 | 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 5 of 5 shared results
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
- ZoomNAS: Searching for Whole-body Human Pose Estimation in the Wild (TPAMI 2022 · similarity 0.95)
- Single-Network Whole-Body Pose Estimation (ICCV 2019 · similarity 0.91)
- H3WB: Human3.6M 3D WholeBody Dataset and Benchmark (ICCV 2023 · similarity 0.91)
- DOPE: Distillation Of Part Experts for whole-body 3D pose estimation in the wild (ECCV 2020 · similarity 0.90)
- Towards Robust and Expressive Whole-body Human Pose and Shape Estimation (NeurIPS 2023 · similarity 0.88)
- PoseBH: Prototypical Multi-Dataset Training Beyond Human Pose Estimation (CVPR 2025 · 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.
