Research map / Label-efficient and robust learning / Semi-supervised learning
A Two-Stream Mutual Attention Network for Semi-Supervised Biomedical Segmentation with Noisy Labels
Read the paperCheck 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 two-stream mutual attention instead of the DenseVoxNet baseline | HVSMR 2016 segmentation · HVSMR 2016 | overall score: -0.161 -> -0.024 | Helped |
| Use TSMAN instead of DenseVoxNet | BRATS 2015 segmentation · BRATS 2015 | mean Dice: 0.728 -> 0.792 | Helped |
| Add loss attention f_att^4 (ablation) | TSMAN without attention · HVSMR 2016 | overall score: -1.167 -> -0.737 | Helped |
| Add spatial attention f_att^3 (ablation) | TSMAN with f_att^4 · HVSMR 2016 | overall score: -0.737 -> -0.601 | Helped |
| Add attention f_att^2 (ablation) | TSMAN with f_att^3 and f_att^4 · HVSMR 2016 | overall score: -0.601 -> -0.576 | Helped |
| Add attention f_att^1 (ablation) | TSMAN with f_att^2 to f_att^4 · HVSMR 2016 | overall score: -0.576 -> -0.561 | Helped |
| Smooth the f_att^4 attention map using a 3×3 kernel with variance 0.5 (ablation) | TSMAN loss attention · HVSMR 2016 | overall score: 0.1 score improvement | Helped |
| Use data distillation with a single base model rather than hierarchical distillation (ablation) | Pseudo-label generation · HVSMR 2016, ξ=30% | overall score: 0.104; 0.163; 0.168 vs 0.191 | Mixed |
| Use model distillation rather than hierarchical distillation (ablation) | Pseudo-label generation · HVSMR 2016, ξ=30% | overall score: 0.184 vs 0.191 | Hurt |
| Use model distillation rather than hierarchical distillation (ablation) | Pseudo-label generation · HVSMR 2016, ξ=80% | overall score: 0.771 vs 0.810 | Hurt |
| Rely on data distillation with a single base model when labeled data is scarce (ablation) | Pseudo-label generation · HVSMR 2016, ξ=30% | overall score: 0.005; 0.154; 0.171 | Mixed |
Closest papers by meaning
- Semi-supervised Medical Image Segmentation through Dual-task Consistency (AAAI 2021 · similarity 0.87)
- Parametric Meets Non-parametric: Bridging Dual Prediction Spaces for Semi-supervised Medical Image Segmentation (NeurIPS 2026 · similarity 0.86)
- Medical Image Segmentation with Minimal Labeling Effort: How Far Can We Push the Limits? (AAAI 2026 · similarity 0.85)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.84)
- Boost the Inference with Co-training: A Depth-guided Mutual Learning Framework for Semi-supervised Medical Polyp Segmentation (CVPR 2025 · similarity 0.84)
- SGTC: Semantic-Guided Triplet Co-training for Sparsely Annotated Semi-Supervised Medical Image Segmentation (AAAI 2025 · similarity 0.84)
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.