Research map / Label-efficient and robust learning / Domain adaptation and generalization
Unsupervised Source-Free Ranking of Biomedical Segmentation Models Under Distribution Shift
Joshua Talks, Kevin Marchesini, Luca Lumetti, Federico Bolelli, Anna Kreshuk
Read the paperCheck 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 |
|---|---|---|---|
| use CMR consistency ranking instead of unsupervised baseline TS | CMR-EI (Gauss) · Mitochondria semantic segmentation | Kendall tau: 0.25 -> 0.77 | Improved |
| use CMR instead of classification-derived supervised metrics like NLEEP | CMR-EI (Gauss) · Nuclei semantic segmentation | Kendall tau: 0.11 -> 0.62 | Improved |
| apply input Gaussian noise vs feature DropOut perturbation (ablation) | CMR-EI · Mitochondria semantic segmentation | Kendall tau: 0.77 -> 0.73 | No clear effect |
| use hard NHD instead of soft EI consistency (ablation) | CMR (Gauss) · ToothFairy2 overall | Kendall tau: 0.71 -> 0.90 | Improved |
| use CMR instead of TS on multi-class 3D segmentation | CMR-NHD (Gamma) · ToothFairy2 overall | Kendall tau: 0.33 -> 1.00 | Improved |
| rank post-UDA models with CMR rather than Transfer Score (Mean Teacher) | CMR-EI (Gauss) · Post-UDA EPFL | Kendall tau: 0.45 -> 0.78 | Improved |
| rank post-UDA models with CMR rather than TS (AdaBN) | CMR-EI (Gauss) · Post-UDA EPFL | Kendall tau: 0.50 -> 0.70 | Improved |
| target VNC after Mean Teacher self-training | CMR-EI (Gauss) · Post-UDA VNC | Kendall tau: 0.78 -> 0.21 | Hurt |
| use CMR-ARS instead of SEG for cell instance ranking | CMR-ARS (DropOut) · Cell instance segmentation | Kendall tau: 0.12 -> 0.69 | Improved |
| use CMR-ARS instead of SEG for nuclei instance ranking | CMR-ARS (DropOut) · Nuclei instance segmentation | Kendall tau: 0.48 -> 0.80 | Improved |
| vary Gaussian noise strength sigma from 0.01-0.03 to 0.10-0.12 (ablation) | CMR-EI (Gauss) · Mitochondria EPFL target | Kendall tau: 0.73 -> 0.87 | No clear effect |
| use a single perturbation per image rather than ~10 runs (ablation) | CMR · semantic segmentation ranking | correlation | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- LEEP — LEEP: A New Measure to Evaluate Transferability of Learned Representations (ICML 2020) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MedSAMix: A Training-Free Model Merging Approach for Medical Image Segmentation (AAAI 2026 · similarity 0.87)
- UniPro: Unified Multi-Mode Medical Image Segmentation from 2D Images to 3D Volumes via Propagation (NeurIPS 2026 · similarity 0.86)
- Versatile Medical Image Segmentation Learned from Multi-Source Datasets via Model Self-Disambiguation (CVPR 2024 · similarity 0.86)
- Why Invariance is Not Enough for Biomedical Domain Generalization and How to Fix It (NeurIPS 2026 · similarity 0.86)
- S2S2: Semantic Stacking for Robust Semantic Segmentation in Medical Imaging (AAAI 2025 · similarity 0.85)
- nnWNet: Rethinking the Use of Transformers in Biomedical Image Segmentation and Calling for a Unified Evaluation Benchmark (CVPR 2025 · 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.
