Found 1 document(s)
Page 1 of 1
-
Realistic few-shot adaptation of medical vision-language models (VLMs) can leave low-prevalence classes without any labeled support. The training-free SS- Text+ solver remains operational in this setting, but its zero-su...