Research map / Label-efficient and robust learning / Semi-supervised learning
FocalMix: Semi-Supervised Learning for 3D Medical Image Detection
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 |
|---|---|---|---|
| Add 400 unlabeled scans to training | 3D FPN with FocalMix · LUNA16, 25 labeled scans | CPM (%): 66.6 -> 78.1; 11.5 (17.3%) | Helped |
| Add 400 unlabeled scans to training | 3D FPN with FocalMix · LUNA16, 50 labeled scans | CPM (%): 75.4 -> 82.0; 6.6 (8.8%) | Helped |
| Add 400 unlabeled scans to training | 3D FPN with FocalMix · LUNA16, 100 labeled scans | CPM (%): 82.8 -> 87.2; 4.4 (5.3%) | Helped |
| Add NLST unlabeled data to FocalMix training | 3D FPN · LUNA16, 533 labeled scans and around 3,000 NLST unlabeled scans | CPM (%): 90.0 -> 90.7 | Helped |
| Add soft α weighting for soft targets (ablation) | Soft-target focal loss in FocalMix · LUNA16, 100 labeled + 400 unlabeled scans | CPM (%): 84.4 -> 85.2 | Helped |
| Add soft β weighting for soft targets (ablation) | Soft-target focal loss in FocalMix · LUNA16, 100 labeled + 400 unlabeled scans | CPM (%): 83.7 -> 85.2 | Helped |
| Use soft-target generalizations for both α and β rather than pseudo-hard targets (ablation) | Soft-target focal loss in FocalMix · LUNA16, 100 labeled + 400 unlabeled scans | training convergence: Fail -> 85.2 | Helped |
| Increase target-prediction augmentations from one to four (ablation) | FocalMix · LUNA16, 100 labeled + 400 unlabeled scans | CPM (%): 85.9 -> 87.2 | Helped |
| Increase target-prediction augmentations from four to eight (ablation) | FocalMix · LUNA16, 100 labeled + 400 unlabeled scans | CPM (%): 87.2 -> 87.1 | No clear effect |
| Add image-level MixUp (ablation) | FocalMix · LUNA16, 100 labeled + 400 unlabeled scans | CPM (%): 85.2 -> 86.7 | Helped |
| Add object-level MixUp after image-level MixUp (ablation) | FocalMix · LUNA16, 100 labeled + 400 unlabeled scans | CPM (%): 86.7 -> 87.2 | Helped |
| Add image-level and object-level MixUp | Fully-supervised 3D FPN · LUNA16, 533 labeled scans | CPM (%): 89.2 -> 90.0 | Helped |
Closest papers by meaning
- Systematic Comparison of Semi-supervised and Self-supervised Learning for Medical Image Classification (CVPR 2024 · similarity 0.84)
- SGTC: Semantic-Guided Triplet Co-training for Sparsely Annotated Semi-Supervised Medical Image Segmentation (AAAI 2025 · similarity 0.83)
- Enhancing vision-language models for medical imaging: bridging the 3D gap with innovative slice selection (NeurIPS 2024 · similarity 0.83)
- Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding (ICLR 2025 · similarity 0.83)
- Autoregressive Sequence Modeling for 3D Medical Image Representation (AAAI 2025 · similarity 0.83)
- BiomedCoOp: Learning to Prompt for Biomedical Vision-Language Models (CVPR 2025 · similarity 0.82)
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.