HERO: Enhancing Multimodal Faithfulness via Dynamic Entropy-Aware Reinforcement Learning

Published in Under Review at ECCV, 2026