Research map / Label-efficient and robust learning / Semi-supervised learning
Bidirectional Copy-Paste for Semi-Supervised Medical Image Segmentation
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 bidirectional labeled-unlabeled copy-paste rather than inward-only, outward-only, or within-set copy-paste (ablation) | BCP · LA and ACDC, 5% labeled | Dice: 87.35 -> 88.02; 87.32 -> 88.02; 79.67 -> 88.02; 81.68 -> 87.59; 72.19 -> 87.59; 81.80 -> 87.59 | Helped |
| Replace BCP with Mixup or FG-CutMix (ablation) | BCP · LA, 5% labeled | Dice: 41.71; 67.05 vs 88.02 | Hurt |
| Use a zero-centered mask rather than random or contact masks (ablation) | BCP · LA, 5% labeled | Dice: 86.06 -> 88.02; 86.64 -> 88.02 | Helped |
| Increase the copied region ratio from beta 1/3 toward beta 2/3 (ablation) | BCP · LA, 5% and 10% labeled | Dice: 79.92 -> 88.02; 83.20 -> 89.62 | Helped |
| Set the mask region ratio to beta 5/6 instead of beta 2/3 (ablation) | BCP · LA, 5% and 10% labeled | Dice: 88.02 -> 87.92; 89.62 -> 88.75 | Hurt |
| Increase the unlabeled-pixel loss weight alpha to 2.5 (ablation) | BCP · LA and ACDC, 5% labeled | Dice: 88.02 -> 86.56; 87.59 -> 85.43 | Hurt |
| Pretrain on labeled data with copy-paste rather than random initialization or pretraining without copy-paste (ablation) | BCP · LA, 5% labeled | Dice: 86.15 -> 88.02; 86.46 -> 88.02 | Helped |
| Add BCP to the pseudo-label self-training baseline (ablation) | BCP · ACDC, 5% labeled | Dice: 47.62 -> 83.26 | Helped |
| Add pseudo-label post-processing and copy-paste pretraining after BCP (ablation) | BCP · ACDC, 5% labeled | Dice: 83.26 -> 87.59 | Helped |
| Use BCP instead of the strongest reported baseline | V-Net / U-Net semi-supervised segmentation · LA 5% and 10%; Pancreas-NIH 20%; ACDC 5% and 10% labeled | Dice: 86.33 -> 88.02; 88.55 -> 89.62; 79.67 -> 82.91; 65.83 -> 87.59; 86.78 -> 88.84 | Helped |
Closest papers by meaning
- Constructing and Exploring Intermediate Domains in Mixed Domain Semi-supervised Medical Image Segmentation (CVPR 2024 · similarity 0.89)
- Adaptive Bidirectional Displacement for Semi-Supervised Medical Image Segmentation (CVPR 2024 · similarity 0.88)
- Semi-supervised TEE Segmentation via Interacting with SAM Equipped with Noise-Resilient Prompting (AAAI 2024 · similarity 0.87)
- Steady Progress Beats Stagnation: Mutual Aid of Foundation and Conventional Models in Mixed Domain Semi-Supervised Medical Image Segmentation (CVPR 2025 · similarity 0.87)
- GapMatch: Bridging Instance and Model Perturbations for Enhanced Semi-Supervised Medical Image Segmentation (AAAI 2025 · similarity 0.87)
- A Semantic Knowledge Complementarity based Decoupling Framework for Semi-supervised Class-imbalanced Medical Image Segmentation (CVPR 2025 · similarity 0.87)
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.