Research map / 3D vision / Human pose estimation
Detection, Pose Estimation and Segmentation for Multiple Bodies: Closing the Virtuous Circle
Miroslav Purkrabek, Jiri Matas
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 mask conditioning (MaskPose) to ViTPose | ViTPose-B · OCHuman test | pose AP: 42.6 -> 45.0 | Improved |
| add second BMP iteration | BBox-Mask-Pose · OCHuman test | pose AP: 46.6 -> 49.3 | Improved |
| add second BMP iteration | BBox-Mask-Pose · COCO val | pose AP: 76.5 -> 76.5 | No clear effect |
| add second BMP iteration | BBox-Mask-Pose · OCHuman test | bbox AP: 32.4 -> 35.9 | Improved |
| add SAM to the loop (versus omitting SAM) (ablation) | BMP 2x · OCHuman val | bbox AP: 30.8 -> 32.1 | Improved |
| add SAM to the loop (versus omitting SAM) (ablation) | BMP 2x · OCHuman val | pose AP: 47.0 -> 48.6 | Improved |
| add keypoint prompts to SAM bounding-box prompt (ablation) | BMP 2x · OCHuman val | bbox AP: 31.9 -> 32.1 | Improved |
| add keypoint prompts to SAM bounding-box prompt (ablation) | BMP 2x · OCHuman val | pose AP: 47.3 -> 48.6 | Improved |
| rerun pose estimation on SAM-refined masks (ablation) | BMP 1x · OCHuman val | pose AP: 45.3 -> 46.4 | Improved |
| use pose-selected instances (c and d) rather than confidence only (ablation) | SAM prompting · OCHuman val | bbox AP: 29.7 -> 32.7 | Improved |
| add negative keypoint prompts with bounding box (ablation) | SAM prompting · OCHuman val | segm AP: 30.9 -> 28.2 | Hurt |
| use batch processing of all instances (ablation) | SAM prompting · OCHuman val | bbox AP: 29.3 -> 29.3 | No clear effect |
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.
- Sapiens 0.3b — Sapiens: Foundation for Human Vision Models (ECCV 2024) · better on 2 of 2 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
- AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking in Real-Time (TPAMI 2022 · similarity 0.86)
- Joint Human Pose Estimation and Instance Segmentation with PosePlusSeg (AAAI 2022 · similarity 0.86)
- Rethinking Pose Estimation in Crowds: Overcoming the Detection Information Bottleneck and Ambiguity (ICCV 2023 · similarity 0.86)
- Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation (AAAI 2020 · similarity 0.86)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.85)
- GigaHumanDet: Exploring Full-Body Detection on Gigapixel-Level Images (AAAI 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.
