Optimizing Vision-Language Consistency via Cross-Layer Regional Attention Alignment

Yifan Wang, Hongfeng Ai, Quangao Liu +7

Vision Language Models (VLMs) face challenges in effectively coordinating diverse attention mechanisms for cross-modal embedding learning, leading to mismatched attention and subop…