A unified 210K medical grounding dataset and full fine-tuning raise a general model's localization F1 from 10.6 to 85.6 on held-out data and improve cross-source transfer.
International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LocAnyMed: Vision-Language Grounding for Multimodal Medical Images
A unified 210K medical grounding dataset and full fine-tuning raise a general model's localization F1 from 10.6 to 85.6 on held-out data and improve cross-source transfer.