VTI-CoT: Visual-Textual Interleaved Chain of Thought for Video Reasoning

Arxiv CS.CV (computer vision)

Introduces VTI-CoT, a method that interleaves visual and textual reasoning chains for improved video understanding. The approach addresses limitations of existing CoT methods by enabling fine-grained cross-modal reasoning across temporal events.

chain-of-thoughtmultimodal

Arxiv CS.CV (computer vision)·Shufan ZhangZiyue LinBairun WangLei JinXuanding DingXinzhu MaKunlin Yang

6 days ago