DualG-MRAG Decouples Macro-Reasoning and Micro-Matching for Multimodal RAG

DualG-MRAG uses a two-tier graph architecture to separate global structural reasoning from fine-grained visual evidence matching in multimodal RAG. This decoupling suppresses retrieval noise and prevents the information loss typically caused by coarse-grained representations in complex multi-hop tasks.