Visually Grounded Self-Reflection for Vision-Language Models via Reinforcement Learning

Large vision-language models can reason over multimodal inputs by generating textual chains of thought (CoT).

Large vision-language models can reason over multimodal inputs by generating textual chains of thought (CoT). A key capability exhibited in CoT reasoning is self-reflection: revisiting earlier decisions and correcting previous errors. However, existing LVLMs often fail to…