Visual CoT: Unleashing Chain-of-Thought Reasoning in Multi-Modal Language Models

A novel pipeline that leverages the reasoning capabilities of multi-modal large language models (MLLMs) by incorporating visual Chain-of-Thought (CoT) reasoning and is capable of evaluating MLLMs in scenarios requiring specific local region identification is presented.