Research map / Detection and segmentation / Weakly supervised detection and segmentation
Point-Teaching: Weakly Semi-supervised Object Detection with Point Annotations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Hungarian point matching instead of no point matching (ablation) | Point-Teaching · COCO val, 1% protocol | AP50:95: 20.2 -> 25.2 | Helped |
| Use randomly sampled points instead of center points (ablation) | Point-Teaching · COCO val, 1% protocol | AP50:95: 25.19 -> 25.18 | No clear effect |
| Add image-wise MIL (ablation) | Point-Teaching with Hungarian matching · COCO val, 1% protocol | AP50:95: 25.2 -> 25.7 | Helped |
| Add point-wise MIL (ablation) | Point-Teaching with Hungarian matching · COCO val, 1% protocol | AP50:95: 25.2 -> 26.0 | Helped |
| Add point-guided copy-paste (ablation) | Point-Teaching with Hungarian matching and both MIL losses · COCO val, 1% protocol | AP50:95: 26.3 -> 27.3 | Helped |
| Set image-wise MIL loss weight to 1.0 rather than 0.5 or 1.5 (ablation) | Point-Teaching · COCO val, 1% protocol | AP50:95: 25.00, 25.66, 25.01 | Mixed |
| Set point-wise MIL loss weight to 0.05 rather than 0.025, 0.1, or 0.15 (ablation) | Point-Teaching · COCO val, 1% protocol | AP50:95: 25.85, 25.97, 25.74, 25.40 | Mixed |
| Set pseudo-box score threshold to 0.05 rather than 0.01, 0.1, or 0.15 (ablation) | Point-Teaching · COCO val, 1% protocol | AP50:95: 26.24, 26.28, 26.17, 26.19 | Mixed |
| Increase point-labeled data while holding fully labeled data at 1% (ablation) | Point-Teaching · COCO val, 1% fully labeled | AP: 9.05 -> 26.28 | Helped |
| Use Point-Teaching rather than Group R-CNN | WSSOD-P · COCO val, 5% fully labeled | AP: 30.1 -> 33.15 | Helped |
| Use Point-Teaching rather than Group R-CNN | WSSOD-P · COCO val, 30% fully labeled | AP: 35.4 -> 38.20 | Helped |
| Use Point-Teaching rather than Unbiased Teacher | Semi-supervised object detection · VOC07 with VOC12 trainval point-labeled data | AP50: 80.51 -> 83.03 | Helped |
Closest papers by meaning
- Points As Queries: Weakly Semi-Supervised Object Detection by Points (CVPR 2021 · similarity 0.94)
- Group R-CNN for Weakly Semi-Supervised Object Detection With Points (CVPR 2022 · similarity 0.92)
- Relational Matching for Weakly Semi-Supervised Oriented Object Detection (CVPR 2024 · similarity 0.91)
- Point2RBox-v2: Rethinking Point-supervised Oriented Object Detection with Spatial Layout Among Instances (CVPR 2025 · similarity 0.88)
- Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.88)
- PointOBB-v2: Towards Simpler, Faster, and Stronger Single Point Supervised Oriented Object Detection (ICLR 2025 · 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.