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

R3G: A Reasoning-Retrieval-Reranking Framework for Vision-Centric Answer Generation arXiv:2602.00104v3 Announce Type: replace Abstract: 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 integrating them effectively into the model's reasoning remains challenging.To address this challenge, we propose R3G, a modular Reasoning-Retrieval-Reranking framework.It first produc