Suresh Taddewadikar

Beyond Visual CoT: Internalized Visual Thinking for Proactive Video Reasoning

Multimodal large language models increasingly use visual chain-of-thought (Visual CoT) to reason about spatial, temporal, and embodied environments. By generating intermediate reasoning images, Visual CoT provides an intuitive mechanism for visual foresight but introduces substantial inference overhead, which is partic...