Research map / Detection and segmentation / Open-set and few-shot detection
Pose2Seg: Detection Free Human Instance Segmentation
Song-Hai Zhang, Ruilong Li, Xin Dong, Paul Rosin, Zixi Cai, Xi Han, Dingcheng Yang, Haozhi Huang, Shi-Min Hu
Check 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 detection-based Mask R-CNN with pose-based Pose2Seg | human instance segmentation · OCHuman val | AP: 0.163 -> 0.222 | Improved |
| Replace detection-based Mask R-CNN with pose-based Pose2Seg | human instance segmentation · OCHuman test | AP: 0.169 -> 0.238 | Improved |
| Replace detection-based Mask R-CNN with pose-based Pose2Seg | human instance segmentation · COCOPerson val | AP: 0.532 -> 0.555 | Improved |
| Use ground-truth rather than predicted keypoints | Pose2Seg · OCHuman val | AP: 0.222 -> 0.544 | Improved |
| Replace GT-box RoI-Align with pose-based Affine-Align (ablation) | human instance segmentation · OCHuman val | AP: 0.476 -> 0.544 | Improved |
| Add skeleton features to GT-box RoI-Align (ablation) | human instance segmentation · COCOPerson val | AP: 0.568 -> 0.648 | Improved |
| Add skeleton features to GT-box RoI-Align (ablation) | human instance segmentation · OCHuman val | AP: 0.133 -> 0.476 | Improved |
| Replace Affine-Align with keypoint-derived boxes and RoI-Align (ablation) | Pose2Seg · OCHuman val | AP: 0.544 -> 0.441 | Hurt |
| Increase SegModule residual units from 5 to 10 (ablation) | Pose2Seg · COCOPerson val | AP: 0.545 -> 0.555 | Improved |
| Increase SegModule residual units beyond 10 (ablation) | Pose2Seg · COCOPerson val | AP: 0.555 -> 0.555; 0.555 -> 0.556 | No clear effect |
| Use keypoint-derived boxes expanded by different amounts instead of Affine-Align (ablation) | Pose2Seg · COCOPerson val | AP: 0.431; 0.460; 0.477; 0.489; 0.497; 0.501; 0.500 | Mixed |
Later papers that compared with this method
2 later accepted papers report Pose2Seg as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Crowd-SAM:SAM as a smart annotator for object detection in crowded scenes (ECCV 2024 · better on 1 of 1 shared result)
- Joint Human Pose Estimation and Instance Segmentation with PosePlusSeg (AAAI 2022 · 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
- Joint Human Pose Estimation and Instance Segmentation with PosePlusSeg (AAAI 2022 · similarity 0.88)
- PersonLab: Person Pose Estimation and Instance Segmentation with a Bottom-Up, Part-Based, Geometric Embedding Model (ECCV 2018 · similarity 0.87)
- Explicit Occlusion Reasoning for Multi-Person 3D Human Pose Estimation (ECCV 2022 · similarity 0.87)
- Joint Multi-Person Pose Estimation and Semantic Part Segmentation (CVPR 2017 · similarity 0.87)
- Occluded Video Instance Segmentation: A Benchmark (IJCV 2022 · similarity 0.86)
- Detangling People: Individuating Multiple Close People and Their Body Parts via Region Assembly (CVPR 2017 · 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.
