Research map / Label-efficient and robust learning / Knowledge distillation
Make a Strong Teacher with Label Assistance: A Novel Knowledge Distillation Approach for Semantic Segmentation
Shoumeng Qiu, Jie Chen, Xinrun Li, Ru Wan, Xiangyang Xue, Jian Pu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 dual-path consistency loss when training the label-assisted teacher (ablation) | LAD teacher training · Cityscapes val, DeepLabV3-R18 student | student mIoU: 76.7 -> 77.7 | Improved |
| add dual-path consistency loss when training the label-assisted teacher (ablation) | LAD teacher training · Cityscapes val, STDC2 student | student mIoU: 73.6 -> 74.8 | Improved |
| feed clean label (no noising) to teacher (ablation) | LAD teacher · Cityscapes val, STDC2 | student mIoU: 74.8 -> 69.4 | Hurt |
| use label-assisted noised-label teacher instead of stronger conventional teachers (ablation) | distillation for PSPNet-R18 · Cityscapes val | student mIoU: 75.9 (PSPNet-R101) -> 76.2 | Improved |
| use label-assisted simple teacher (STDC1) for complex student (STDC2) (ablation) | distillation across complexity · Cityscapes val | student mIoU: 73.2 -> 74.6 | Improved |
| increase pixel-wise label noise from 1e-2 to 1.0 (ablation) | label noising · Cityscapes val, STDC2 | student mIoU: 74.8 -> 70.5 | Hurt |
| reduce pixel-wise noise to near zero with class-wise noising (ablation) | label noising · Cityscapes val, STDC2 | student mIoU: 74.8 -> 73.8 | Hurt |
| apply LAD over CWD baseline on COCO-Stuff 10K with DeepLabV3-R50 | CWD distillation · COCO-Stuff 10K | mIoU: 35.5 -> 35.6 | No clear effect |
| apply LAD over CWD baseline on FCN-50 | CWD distillation · Cityscapes val | mIoU: 74.4 -> 74.5 | No clear effect |
| apply LAD to OCR-W18 distillation over CWD | CWD distillation · Cityscapes val | mIoU: 76.9 -> 78.1 | Improved |
Closest papers by meaning
- Structured Knowledge Distillation for Semantic Segmentation (CVPR 2019 · similarity 0.91)
- Distilling Knowledge from Heterogeneous Architectures for Semantic Segmentation (AAAI 2025 · similarity 0.91)
- Adaptive Perspective Distillation for Semantic Segmentation (TPAMI 2022 · similarity 0.91)
- Label-Guided Knowledge Distillation for Continual Semantic Segmentation on 2D Images and 3D Point Clouds (ICCV 2023 · similarity 0.91)
- Knowledge Distillation from A Stronger Teacher (NeurIPS 2022 · similarity 0.90)
- Self-Decoupling and Ensemble Distillation for Efficient Segmentation (AAAI 2023 · 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.
