Preprint
Aug 2026
MVC-Bench: Benchmarking Calibration of Medical Vision-Language Models
The underlying causes of miscalibration in VLMs and Medical-VLMs are investigated and a simple train-time calibration method, Multi-Class Margin (MCM) regularization, is proposed, which achieves lowest ECE on 10 out of 12 settings in in-domain and remains competitive under domain shifts.
Ashshak Sharifdeen, Shihab Aaqil Ahamed, Ufaq Khan et al.
· 2 citations