Research map / Label-efficient and robust learning / Semi-supervised learning
Perturbed and Strict Mean Teachers for Semi-Supervised Semantic Segmentation
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 the proposed method instead of CPS | DeepLabV3+ on Pascal VOC 2012 · Pascal VOC 2012 labelled partitions, ResNet-50 and ResNet-101 | mIoU: 72.83 vs 71.98; 75.70 vs 73.67; 76.43 vs 74.90; 77.88 vs 76.15 (ResNet-50); 75.50 vs 74.48; 78.20 vs 76.44; 78.72 vs 77. | Helped |
| Use the proposed method instead of CPS in the matched setting | DeepLabV3+ ResNet-50 on Cityscapes · Cityscapes labelled partitions 1/8, 1/4, and 1/2 | mIoU: 77.12 vs 76.61; 78.38 vs 77.83; 79.22 vs 78.77 | Helped |
| Replace MSE consistency loss with confidence-weighted CE (ablation) | Mean Teacher on Pascal VOC 2012 · Pascal VOC 2012 1/8, ResNet-50 and ResNet-101 | mIoU: +2.30%; +2.81% | Helped |
| Add teacher-estimated adversarial feature perturbation (T-VAT) (ablation) | MT + Conf-CE · Pascal VOC 2012 1/8, ResNet-50 and ResNet-101 | mIoU: 74.87 vs 73.79; 77.36 vs 76.39 | Helped |
| Add the auxiliary teacher (ablation) | MT + Conf-CE + T-VAT · Pascal VOC 2012 1/8, ResNet-50 and ResNet-101 | mIoU: +0.83%; +0.84% | Helped |
| Use T-VAT rather than uniform feature noise, student VAT, or no feature perturbation (ablation) | Proposed method · Pascal VOC 2012 ablation | mIoU: +0.93%; +0.62%; +1.10% | Helped |
| Apply CutMix after prediction rather than before prediction (ablation) | Consistency training · Pascal VOC 2012 validation | mIoU: around 3% | Helped |
| Add the CAM pseudo-label loss to the consistency loss (ablation) | Proposed method on Pascal VOC 2012 · Official labelled set, DeepLabV3+ ResNet-50 and ResNet-101 | mIoU: +0.63%; +1.18% | Helped |
Closest papers by meaning
- Semi-Supervised Semantic Segmentation With Cross-Consistency Training (CVPR 2020 · similarity 0.92)
- Towards the Uncharted: Density-Descending Feature Perturbation for Semi-supervised Semantic Segmentation (CVPR 2024 · similarity 0.91)
- Semi-Supervised Semantic Segmentation With Cross Pseudo Supervision (CVPR 2021 · similarity 0.91)
- Improving Semi-Supervised Semantic Segmentation with Sliced-Wasserstein Feature Alignment and Uniformity (CVPR 2025 · similarity 0.89)
- ScaleMatch: Multi-scale Consistency Enhancement for Semi-supervised Semantic Segmentation (AAAI 2025 · similarity 0.88)
- Relaxed Class-consensus Consistency for Semi-supervised Semantic Segmentation (AAAI 2025 · 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.