Research map / Label-efficient and robust learning / Semi-supervised learning
Diverse Cotraining Makes Strong Semi-Supervised Segmentor
Yijiang Li, Xinjiang Wang, Lihe Yang, Litong Feng, Wayne Zhang, Ying Gao
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 strong-weak augmentation to co-training (ablation) | co-training · Pascal VOC 2012 (1/32, ResNet50) | mIoU: 65.66 -> 70.28 | Improved |
| add strong augmentation to shared-backbone co-training (ablation) | co-training with shared backbone · Pascal VOC 2012 (1/32, ResNet50) | mIoU: 58.97 -> 69.48 | Improved |
| use separate backbones instead of a shared backbone (ablation) | co-training · Pascal VOC 2012 (1/32, ResNet50, w/o SA) | mIoU: 58.97 -> 65.66 | Improved |
| use 3-cps (three models) instead of two models (ablation) | single-architecture co-training · Pascal VOC 2012 (1/32, ResNet50, w/o SA) | mIoU: 65.66 -> 65.41 | No clear effect |
| add DCT frequency-domain model alongside RGB model (ablation) | co-training · Pascal VOC 2012 (1/32, ResNet50, w/o SA) | mIoU: 65.66 -> 69.45 | Improved |
| use HSV instead of DCT as second domain (ablation) | co-training with SA · Pascal VOC 2012 (1/16, ResNet50) | mIoU: 73.36 -> 72.64 | Hurt |
| pair ResNet50 with SegFormer-b2 (different architecture) (ablation) | co-training · Pascal VOC 2012 (1/32, w/o SA) | mIoU: 65.66 -> 71.58 | Improved |
| pair ResNet50 with SegFormer-b2 instead of ResNeSt50 (ablation) | co-training with SA · Pascal VOC 2012 (1/32) | mIoU: 72.70 -> 74.85 | Improved |
| add different architectures on top of strong augmentation (ablation) | Diverse Co-training · Pascal VOC 2012 blended (1/32) | mIoU: 70.28 -> 74.89 | Improved |
| add different domains and different architectures together (ablation) | Diverse Co-training · Pascal VOC 2012 blended (1/32) | mIoU: 74.89 -> 75.21 | Improved |
| add different strong augmentation per model on top of SA (ablation) | co-training · Pascal VOC 2012 blended (1/32) | mIoU: 70.28 -> 71.07 | Improved |
| replace knowledge distillation from SegFormer with mutual co-training (ablation) | FixMatch ResNet50 · Pascal VOC 2012 (1/32) | mIoU: 74.1 -> 75.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FixMatch — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Semi-Supervised Segmentation based on Error-Correcting Supervision (ECCV 2020 · similarity 0.88)
- Conflict-Based Cross-View Consistency for Semi-Supervised Semantic Segmentation (CVPR 2023 · similarity 0.87)
- Semi-supervised Semantic Segmentation via Strong-weak Dual-branch Network (ECCV 2020 · similarity 0.87)
- MultiSeg: Semantically Meaningful, Scale-Diverse Segmentations From Minimal User Input (ICCV 2019 · similarity 0.86)
- Semi-supervised Learning with Multi-Head Co-Training (AAAI 2022 · similarity 0.86)
- Deep Co-Training for Semi-Supervised Image Recognition (ECCV 2018 · similarity 0.86)
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.
