Research map / Detection and segmentation / Weakly supervised detection and segmentation
Unbiased Teacher 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 |
|---|---|---|---|
| use teacher-student mutual learning with EMA and Focal loss | Faster R-CNN FPN ResNet-50 · COCO-standard 1% labeled | mAP: 9.05 ± 0.16 -> 20.75 ± 0.12 | Helped |
| use teacher-student mutual learning with EMA and Focal loss | Faster R-CNN FPN ResNet-50 · COCO-additional | AP50:95: 40.20 -> 41.30 | Helped |
| add EMA teacher updates rather than shared Teacher and Student weights (ablation) | Unbiased Teacher · COCO-standard 1% labeled | mAP: 16.91 -> 21.19 | Helped |
| add EMA teacher updates rather than shared Teacher and Student weights (ablation) | Unbiased Teacher · COCO-standard 1% labeled | KL divergence of pseudo-label distribution: 1.7915 -> 0.2482 | Helped |
| replace cross-entropy with Focal loss (ablation) | Unbiased Teacher without EMA · COCO-standard 1% labeled | mAP: 13.42 -> 17.85 | Helped |
| replace cross-entropy with Focal loss (ablation) | Unbiased Teacher with EMA · COCO-standard 1% labeled | mAP: 16.91 -> 21.19 | Helped |
| add the Burn-In initialization stage (ablation) | Unbiased Teacher · COCO-standard 1% labeled | mAP | Helped |
| filter low-confidence boxes with confidence thresholding (ablation) | Unbiased Teacher pseudo-labels · COCO-standard | pseudo-label accuracy | Helped |
| raise the pseudo-label threshold excessively to 0.9 (ablation) | Unbiased Teacher · COCO-standard | AP | Hurt |
| lower the pseudo-label threshold to 0.6 (ablation) | Unbiased Teacher · COCO-standard | AP | Hurt |
| use EMA rate 0.99 rather than 0.5 (ablation) | Unbiased Teacher · COCO-standard 1% labeled | mAP | Helped |
| increase EMA rate beyond 0.99 to 0.9999 (ablation) | Unbiased Teacher · COCO-standard 1% labeled | mAP | Hurt |
Closest papers by meaning
- Humble Teachers Teach Better Students for Semi-Supervised Object Detection (CVPR 2021 · similarity 0.94)
- Instant-Teaching: An End-to-End Semi-Supervised Object Detection Framework (CVPR 2021 · similarity 0.94)
- Unbiased Teacher v2: Semi-Supervised Object Detection for Anchor-Free and Anchor-Based Detectors (CVPR 2022 · similarity 0.93)
- De-biased Teacher: Rethinking IoU Matching for Semi-supervised Object Detection (AAAI 2023 · similarity 0.92)
- Active Teacher for Semi-Supervised Object Detection (CVPR 2022 · similarity 0.92)
- Relational Matching for Weakly Semi-Supervised Oriented Object Detection (CVPR 2024 · 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.