Research map / Label-efficient and robust learning / Semi-supervised learning
ATSO: Asynchronous Teacher-Student Optimization for Semi-Supervised Image Segmentation
Xinyue Huo, Lingxi Xie, Jianzhong He, Zijie Yang, Wengang Zhou, Houqiang Li, Qi Tian
Check 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 |
|---|---|---|---|
| Train each generation from scratch instead of initializing it from the previous teacher (ablation) | Teacher-student optimization on NIH pancreas · NIH pancreas, 10% labeled, test set | DSC (%): 78.98 -> 79.67 | Improved |
| Use asynchronous reference subsets so a teacher does not label data used to train its student (ablation) | STSO on NIH pancreas · NIH pancreas, 10% labeled, test set | DSC (%): 79.67 -> 81.69 | Improved |
| Use asynchronous reference subsets so a teacher does not label data used to train its student (ablation) | STSO on NIH pancreas · NIH pancreas, 20% labeled, test set | DSC (%): 83.21 -> 83.70 | Improved |
| Use asynchronous reference subsets so a teacher does not label data used to train its student (ablation) | STSO on NIH-to-MSD pancreas transfer · MSD pancreas test set | DSC (%): 75.48 -> 76.71 | Improved |
| Use asynchronous reference subsets rather than synchronous optimization (ablation) | Teacher-student transfer from Cityscapes to Mapillary · Mapillary test, 19 classes | mIoU (%): 28.11 -> 28.26 | No clear effect |
| Reduce pseudo-label classes from 19 to 5 (ablation) | ATSO transfer from Cityscapes to Mapillary · Mapillary test, 5 classes | mIoU (%): 69.94 -> 77.11 | Improved |
| Use 5-class pseudo labels from ATSO rather than 5-class pseudo labels from STSO (ablation) | Target-network training on Mapillary · Mapillary test, 5-class pseudo-label supervision | mIoU (%): 29.76 -> 34.36 | Improved |
| Tune the 5-class ATSO model using 19-class pseudo labels (ablation) | ATSO transfer from Cityscapes to Mapillary · Mapillary test, 19 classes | mIoU (%): 34.36 -> 37.23 | Improved |
| Apply self-learning with the previous generation as initialization instead of training from scratch (ablation) | NIH pancreas semi-supervised learning · NIH pancreas, 10% labeled, test set | DSC (%): 78.98 at G2; 77.78 at G4 | Hurt |
| Use asynchronous optimization instead of synchronous optimization (ablation) | Teacher-student optimization on Cityscapes · Cityscapes, 1K labeled, test set | mIoU (%): 68.72 -> 70.43 | Improved |
Closest papers by meaning
- Semi-supervised TEE Segmentation via Interacting with SAM Equipped with Noise-Resilient Prompting (AAAI 2024 · similarity 0.86)
- Enhancing SAM with Efficient Prompting and Preference Optimization for Semi-supervised Medical Image Segmentation (CVPR 2025 · similarity 0.86)
- Improving the Generalization of Segmentation Foundation Model under Distribution Shift via Weakly Supervised Adaptation (CVPR 2024 · similarity 0.85)
- GapMatch: Bridging Instance and Model Perturbations for Enhanced Semi-Supervised Medical Image Segmentation (AAAI 2025 · similarity 0.85)
- RWMS: Reliable Weighted Multi-Phase for Semi-supervised Segmentation (AAAI 2024 · similarity 0.85)
- Combinatorial CNN-Transformer Learning with Manifold Constraints for Semi-supervised Medical Image Segmentation (AAAI 2024 · similarity 0.85)
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.
