Applying GRPO reinforcement learning with spatial-semantic rewards and a Chain-of-Box prompt achieves state-of-the-art medical image grounding without chain-of-thought annotations.
In: Proceedings of the Computer Vision and Pattern Recognition Conference
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
MedGround-R1: Advancing Medical Image Grounding via Spatial-Semantic Rewarded Group Relative Policy Optimization
Applying GRPO reinforcement learning with spatial-semantic rewards and a Chain-of-Box prompt achieves state-of-the-art medical image grounding without chain-of-thought annotations.