Research map / Detection and segmentation / Weakly supervised detection and segmentation
Interactive Self-Training With Mean Teachers for Semi-Supervised Object Detection
Check 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 |
|---|---|---|---|
| add interactive self-training with two ROI heads predicting pseudo-labels for each other (ablation) | ISMT · 1% and 10% COCO | mAP: 15.94 -> 18.88; 26.75 -> 30.53 | Helped |
| add mean teacher modules to provide cross-head pseudo-labels (ablation) | ISMT · 1% and 10% COCO | mAP: 15.23 -> 18.88; 28.33 -> 30.53 | Helped |
| add NMS fusion of current detections with historical pseudo-labels (ablation) | ISMT · 1% and 10% COCO | mAP: 14.74 -> 18.88; 26.65 -> 30.53 | Helped |
| add DropBlock to ROI heads (ablation) | ISMT · 1% and 10% COCO | mAP: 17.32 -> 18.88; 29.48 -> 30.53 | Helped |
| use two ROI heads rather than one (ablation) | ISMT · 1% and 10% COCO | mAP: 15.64 -> 18.88; 26.47 -> 30.53 | Helped |
| increase the number of ROI heads from two to three (ablation) | ISMT · 1% and 10% COCO | mAP: 18.88 -> 18.53; 30.53 -> 30.74 | Mixed |
| set unlabeled loss weight γ to 0.5 rather than 2 (ablation) | ISMT · 10% COCO | mAP: 24.82 vs 30.53 | Hurt |
| set unlabeled loss weight γ to 1 rather than 2 (ablation) | ISMT · 10% COCO | mAP: 29.42 vs 30.53 | Hurt |
| set unlabeled loss weight γ to 4 rather than 2 (ablation) | ISMT · 10% COCO | mAP: 28.94 vs 30.53 | Hurt |
| set unlabeled loss weight γ to 8 rather than 2 (ablation) | ISMT · 10% COCO | mAP: 18.10 vs 30.53 | Hurt |
| set EMA smoothing coefficient α to 0.5 or 0.9 rather than 0.99 (ablation) | ISMT · 10% COCO | mAP: 27.82, 28.42 vs 30.53 | Hurt |
| set EMA smoothing coefficient α to 0.999 or 0.9999 rather than 0.99 (ablation) | ISMT · 10% COCO | mAP: 29.94, 26.10 vs 30.53 | Hurt |
Closest papers by meaning
- Instant-Teaching: An End-to-End Semi-Supervised Object Detection Framework (CVPR 2021 · similarity 0.91)
- Humble Teachers Teach Better Students for Semi-Supervised Object Detection (CVPR 2021 · similarity 0.91)
- Label Matching Semi-Supervised Object Detection (CVPR 2022 · similarity 0.91)
- Unbiased Teacher for Semi-Supervised Object Detection (ICLR 2021 · similarity 0.90)
- Active Teacher for Semi-Supervised Object Detection (CVPR 2022 · similarity 0.90)
- Sparse Semi-DETR: Sparse Learnable Queries for Semi-Supervised 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.