From Scenes to Elements: Multi-Granularity Evidence Retrieval for Verifiable Multimodal RAG

TL;DR

This paper introduces a multi-granularity evidence retrieval framework for multimodal RAG systems, improving fine-grained, verifiable evidence retrieval using element-level annotations and alignment.

Contribution

It presents GranuRAG, a novel multi-stage framework that treats visual elements as first-class retrieval units, enhancing transparency and accuracy in multimodal evidence retrieval.

Findings

GranuRAG achieves up to 29.2% improvement over baseline methods.

The benchmark GranuVistaVQA captures partial observation challenges with element-level annotations.

Grounding retrieval at the element level enables better error diagnosis.

Abstract

Multimodal Retrieval-Augmented Generation (RAG) systems retrieve evidence at coarse granularities (entire images or scenes), creating a mismatch with fine-grained user queries and making failures unverifiable. We introduce GranuVistaVQA, a multimodal benchmark featuring real-world landmarks with element-level annotations across multiple viewpoints, capturing the partial observation challenge where individual images contain only subsets of entities. We further propose GranuRAG, a multi-granularity framework that treats visual elements as first-class retrieval units through three stages: element-level detection and classification, multi-granularity cross-modal alignment for evidence retrieval, and attribution-constrained generation. By grounding retrieval at the element level rather than relying on implicit attention, our approach enables transparent error diagnosis. Experiments demonstrate that GranuRAG achieves up to 29.2% improvement over six strong baselines for this task.