R3G: A Reasoning-Retrieval-Reranking Framework for Vision-Centric Answer Generation

Zhuohong Chen, Zhengxian Wu, Zirui Liao +6

Vision-centric retrieval for VQA requires retrieving images to supply missing visual cues and integrating them into the reasoning process. However, selecting the right images and i…