Research map / Detection and segmentation / Weakly supervised detection and segmentation
Object Localization Under Single Coarse Point Supervision
Xuehui Yu, Pengfei Chen, Di Wu, Najmul Hassan, Guorong Li, Junchi Yan, Humphrey Shi, Qixiang Ye, Zhenjun Han
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 |
|---|---|---|---|
| refine coarse points with CPR before P2PNet localization | P2PNet · COCO, DOTA, SeaPerson | mAP^all_1.0: 38.48 -> 55.46; 48.34 -> 63.81; 76.52 -> 85.86 | Improved |
| add MIL loss to annotation and negative losses (ablation) | CPR + P2PNet · COCO | mAP^all_1.0: 51.82 -> 55.46 | Improved |
| replace MIL loss with positive-point loss (ablation) | CPR + P2PNet · COCO | mAP^all_1.0: 55.46 -> 42.72 | Hurt |
| add annotation loss to MIL and negative losses (ablation) | CPR + P2PNet · COCO | mAP^all_1.0: 54.24 -> 55.46 | Improved |
| add negative loss to MIL and annotation losses (ablation) | CPR + P2PNet · COCO | mAP^all_1.0: 39.45 -> 55.46 | Improved |
| use P3 sampling radius R=8 rather than R=5 (ablation) | CPR · COCO | mAP^all_1.0: 53.32 -> 55.46 | Improved |
| increase P3 sampling radius from R=8 to R=25 (ablation) | CPR · COCO | mAP^all_1.0: 55.46 -> 53.85 | Hurt |
| retain all three point-refinement constraints (ablation) | CPR · COCO | mAP^all_1.0: 54.25 -> 55.46; 52.69 -> 55.46 | Improved |
| use coarse point annotations with CPR instead of coarse points without CPR (ablation) | P2PNet · COCO | mAP^all_1.0: 38.48 -> 55.46 | Improved |
| use coarse points with CPR rather than center-point annotations (ablation) | P2PNet · COCO | mAP^all_1.0: 55.46 -> 57.47 | Hurt |
| use ResNet-101 instead of ResNet-50 for CPRNet and P2PNet (ablation) | CPR + P2PNet · COCO | mAP^all_1.0: 55.46 -> 56.43 | Improved |
| replace circular point sampling with rectangular sampling (ablation) | CPR · COCO | mAP^all_1.0: 55.46 -> 55.39 | 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.
- Faster R-CNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · 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
- CPR++: Object Localization via Single Coarse Point Supervision (TPAMI 2024 · similarity 0.95)
- Unsupervised Object Localization with Representer Point Selection (ICCV 2023 · similarity 0.87)
- RepPoints: Point Set Representation for Object Detection (ICCV 2019 · similarity 0.86)
- A Coarse-Fine Network for Keypoint Localization (ICCV 2017 · similarity 0.85)
- Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.84)
- PointSR: Self-Regularized Point Supervision for Drone-View Object Detection (CVPR 2025 · 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.
