Research map / Detection and segmentation / Weakly supervised detection and segmentation
Humble Teachers Teach Better Students 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 |
|---|---|---|---|
| Use Humble Teacher instead of STAC | Faster R-CNN semi-supervised detection · VOC07 test (VOC07 labeled, VOC12 unlabeled) | AP: 44.64 -> 53.04 | Helped |
| Use Humble Teacher instead of STAC | Faster R-CNN, ResNet-50 · MS-COCO val 2017 (1%, 2%, 5%, and 10% labeled) | mAP (50:95): 13.97±0.35 -> 16.96±0.38; 18.25±0.25 -> 21.72±0.24; 24.38±0.12 -> 27.70±0.15; 28.64±0.21 -> 31.61±0.28 | Helped |
| Add unlabeled MS-COCO data with Humble Teacher | Faster R-CNN, ResNet-50 · MS-COCO val 2017 (full train labeled + unlabeled) | AP: 37.63 -> 42.37 | Helped |
| Add unlabeled MS-COCO data with Humble Teacher | Cascade R-CNN, ResNet-152 · MS-COCO test-dev 2017 | AP: 50.7 -> 53.8 | Helped |
| Update teacher weights with EMA at every iteration rather than keeping the teacher fixed (ablation) | Humble Teacher · MS-COCO train 2017 (10% labeled) | AP: 27.26±0.21 -> 31.61±0.28 | Helped |
| Update teacher weights from the student every 10K iterations rather than keeping the teacher fixed (ablation) | Humble Teacher · MS-COCO train 2017 (10% labeled) | AP: 27.26±0.21 -> 28.61±0.18 | Helped |
| Use per-iteration EMA rather than copying student weights every 10K iterations (ablation) | Humble Teacher · MS-COCO train 2017 (10% labeled) | AP: 28.61±0.18 -> 31.61±0.28 | Helped |
| Use soft labels instead of confidence-thresholded hard labels (ablation) | Humble Teacher · MS-COCO train 2017 (10% labeled) | AP: 27.97±0.13 -> 30.97±0.16 | Helped |
| Add a task-specific ensemble of original and horizontally flipped images (ablation) | Humble Teacher · MS-COCO train 2017 (10% labeled) | AP: 30.97±0.16 -> 31.61±0.28 | Helped |
| Use a random augmented teacher ensemble instead of no ensemble (ablation) | Humble Teacher · MS-COCO train 2017 (10% labeled) | AP: 30.97±0.16 -> 30.79±0.31 | Hurt |
| Vary the number of teacher proposals used for the unsupervised ROI loss (ablation) | Humble Teacher · MS-COCO | AP | Mixed |
Closest papers by meaning
- Unbiased Teacher for Semi-Supervised Object Detection (ICLR 2021 · similarity 0.94)
- Active Teacher for Semi-Supervised Object Detection (CVPR 2022 · similarity 0.93)
- Instant-Teaching: An End-to-End Semi-Supervised Object Detection Framework (CVPR 2021 · similarity 0.93)
- DTG-SSOD: Dense Teacher Guidance for Semi-Supervised Object Detection (NeurIPS 2022 · similarity 0.92)
- Point-Teaching: Weakly Semi-supervised Object Detection with Point Annotations (AAAI 2023 · similarity 0.91)
- A-Teacher: Asymmetric Network for 3D Semi-Supervised Object Detection (CVPR 2024 · similarity 0.90)
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.