Research map / Detection and segmentation / Weakly supervised detection and segmentation
De-biased Teacher: Rethinking IoU Matching for Semi-supervised Object Detection
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 |
|---|---|---|---|
| Replace pseudo-label IoU matching with direct soft-labeled proposals and weak/strong consistency regularization | De-biased Teacher · COCO train2017 10% labeled / COCO-val2017 | mAP: 35.02±0.26 -> 35.50±0.20 | Helped |
| Replace pseudo-label IoU matching with direct soft-labeled proposals and weak/strong consistency regularization | De-biased Teacher · PASCAL VOC2007-test | AP50: 80.94 -> 81.50 | Helped |
| Replace pseudo-label IoU matching with direct soft-labeled proposals and weak/strong consistency regularization | De-biased Teacher · Open scene Object365 validation | mAP: 25.0(+1.9) -> 28.1 | Helped |
| Replace pseudo-label IoU matching with direct soft-labeled proposals and weak/strong consistency regularization | De-biased Teacher · Open scene COCO-val2017 | mAP: 42.8 -> 43.1 | Helped |
| Add distribution refinement that truncates low-probability classes and renormalizes target distributions (ablation) | De-biased Teacher · COCO 10% labeled / COCO-val2017 | mAP: +0.3 mAP | Helped |
| Set foreground and refinement threshold δ to 0.01 instead of 0.05 (ablation) | De-biased Teacher · COCO 10% labeled / COCO-val2017 | mAP: 35.5 -> 35.1 | Hurt |
| Set foreground and refinement threshold δ to 0.1 instead of 0.05 (ablation) | De-biased Teacher · COCO 10% labeled / COCO-val2017 | mAP: 35.5 -> 35.4 | Hurt |
| Remove IoU matching and select foreground training boxes directly | De-biased Teacher · COCO 10% labeled, random 8000 images | foreground recall: 0.467 -> 0.782 | Helped |
| Generate foreground training boxes by IoU matching rather than direct selection (ablation) | Soft Teacher training-box analysis · COCO 10% labeled, random 8000 images | precision: 0.892 -> 0.775 | Hurt |
Closest papers by meaning
- Unbiased Teacher for Semi-Supervised Object Detection (ICLR 2021 · similarity 0.92)
- Label Matching Semi-Supervised Object Detection (CVPR 2022 · similarity 0.91)
- Adapting Object Size Variance and Class Imbalance for Semi-supervised Object Detection (AAAI 2023 · similarity 0.90)
- Semi-DETR: Semi-Supervised Object Detection With Detection Transformers (CVPR 2023 · similarity 0.90)
- Relational Matching for Weakly Semi-Supervised Oriented Object Detection (CVPR 2024 · similarity 0.88)
- Sparse Semi-DETR: Sparse Learnable Queries for Semi-Supervised Object Detection (CVPR 2024 · similarity 0.87)
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.