Wasserstein-Aligned Localisation for VLM-Based Distributional OOD Detection in Medical Imaging
Bernhard Kainz, Johanna P Mueller, Matthew Baugh, Cosmin Bercea · 2026
Read Wasserstein-Aligned Localisation for VLM-Based Distributional OOD Detection in Medical Imaging by Bernhard Kainz, Johanna P Mueller, Matthew Baugh, Cosmin Bercea (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-05161.txt.
Description
Bernhard Kainz and three co-authors reframe spotting anomalies in medical scans with vision-language models as a comparison problem rather than a lone-image judgement. Zero-shot localisation is limited because the model lacks healthy anatomical context. Their training-free framework, WALDO, compares a scan against reference distributions of normal anatomy using optimal transport measures.
Reference choice matters in a non-obvious way: moderately similar references work best, a Goldilocks effect the authors analyse theoretically. Similarity-weighted distance measures pick references, and self-consistency steps merge the results. On a brain MRI benchmark the method improved localisation substantially over zero-shot baselines, with consistent gains also seen across other large vision-language models.
Catalog record
| CREATOR | Bernhard Kainz, Johanna P Mueller, Matthew Baugh, Cosmin Bercea |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.05161