1 paper
Shijia Yang, Yunong Liu, Bohan Zhai +5
Image captions serve as efficient surrogates for visual content in multimodal systems such as retrieval, recommendation, and multi-step agentic inference pipelines. Yet current eva…