TCLA

Researchers introduced TCLA, a training-free and model-agnostic approach to improve medical vision-language models' performance on out-of-distribution data by correcting inference logits with few supp…