Research map / Detection and segmentation / Weakly supervised detection and segmentation
Every Annotation Counts: Multi-Label Deep Supervision for Medical Image Segmentation
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 multi-label deep supervision to the mask-supervised UNet | UNet mask segmentation · RETOUCH Spectralis, 3–24 masks and full access | mIoU: +3.18%, +5.94%, +7.57%, +4.05%, +3.73% | Helped |
| Use Mean-Taught instead of Self-Taught pseudo-labeling | UNet with mask supervision and unlabeled images · RETOUCH Spectralis, 6, 12, and 24 masks | mIoU: 28.62 -> 35.17; 43.57 -> 53.52; 56.11 -> 58.84 | Helped |
| Use Mean-Taught rather than Self-Taught deep supervision | UNet with bounding-box supervision and unlabeled images · RETOUCH Spectralis, 24 boxes | mIoU: 32.71 -> 34.14 | Helped |
| Add stronger photometric perturbations to teacher input (ablation) | Mean-Taught Deep Supervision · RETOUCH Spectralis validation, 24 masks plus image-level labels | mIoU: 58.26 -> 61.36 | Helped |
| Use the teacher rather than student predictions for inference (ablation) | Mean-Taught Deep Supervision · RETOUCH Spectralis validation, 24 masks plus image-level labels | mIoU: 58.54 -> 61.36 | Helped |
| Set mean-teacher smoothing coefficient α to 0.5 rather than 0.1 (ablation) | Mean-Taught Deep Supervision · RETOUCH Spectralis validation, 24 masks plus image-level labels | mIoU: 60.15 -> 61.36 | Helped |
| Remove MSE consistency loss (ablation) | Mean-Taught Deep Supervision · RETOUCH Spectralis validation, 24 masks plus image-level labels | mIoU: 61.36 -> 61.24 | No clear effect |
| Add mean-teacher pseudo-labeling to Self-Taught Deep Supervision (ablation) | UNet with 24 pixel masks and remaining image-level labels · RETOUCH Spectralis validation | mIoU: 57.80 -> 61.24 | Helped |
| Add mean-teacher pseudo-labeling with stronger perturbations and teacher inference | UNet with image-level supervision · Cirrus, 24 masks | mIoU: 33.79 -> 35.82 | Helped |
| Add mean-teacher pseudo-labeling with stronger perturbations and teacher inference | UNet with image-level supervision · Topcon, 12 and 24 masks | mIoU: 21.88 -> 23.92; 27.67 -> 33.87 | Helped |
Closest papers by meaning
- Reference-Guided Pseudo-Label Generation for Medical Semantic Segmentation (AAAI 2022 · similarity 0.91)
- Label-Efficient Hybrid-Supervised Learning for Medical Image Segmentation (AAAI 2022 · similarity 0.90)
- Rethinking Box Supervision: Bias-Free Weakly Supervised Medical Segmentation (CVPR 2026 · similarity 0.89)
- Annotation Ambiguity Aware Semi-Supervised Medical Image Segmentation (CVPR 2025 · similarity 0.87)
- Pancakes: Consistent Multi-Protocol Image Segmentation Across Biomedical Domains (NeurIPS 2025 · similarity 0.86)
- SGTC: Semantic-Guided Triplet Co-training for Sparsely Annotated Semi-Supervised Medical Image Segmentation (AAAI 2025 · 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.