Mitigating Cross-Image Information Leakage in Multi-Image Understanding with Large Vision-Language Models
It is demonstrated that FOCUS offers a general solution for enhancing multi-image reasoning without additional training or architectural modifications, and that FOCUS generalizes to video understanding, extending its applicability beyond static multi-image inputs.
Yeji Park, Minyoung Lee, Sanghyuk Chun et al.
· 0 citations