ArtRAG: Retrieval-Augmented Generation with Structured Context for Visual Art Understanding
This addresses the need for culturally and historically informed art explanations, offering a domain-specific improvement over general multimodal models.
The paper tackles the problem of nuanced visual art understanding by proposing ArtRAG, a training-free framework that combines structured knowledge with retrieval-augmented generation, resulting in outperformance on datasets like SemArt and Artpedia and human-evaluated coherent, insightful interpretations.
Understanding visual art requires reasoning across multiple perspectives -- cultural, historical, and stylistic -- beyond mere object recognition. While recent multimodal large language models (MLLMs) perform well on general image captioning, they often fail to capture the nuanced interpretations that fine art demands. We propose ArtRAG, a novel, training-free framework that combines structured knowledge with retrieval-augmented generation (RAG) for multi-perspective artwork explanation. ArtRAG automatically constructs an Art Context Knowledge Graph (ACKG) from domain-specific textual sources, organizing entities such as artists, movements, themes, and historical events into a rich, interpretable graph. At inference time, a multi-granular structured retriever selects semantically and topologically relevant subgraphs to guide generation. This enables MLLMs to produce contextually grounded, culturally informed art descriptions. Experiments on the SemArt and Artpedia datasets show that ArtRAG outperforms several heavily trained baselines. Human evaluations further confirm that ArtRAG generates coherent, insightful, and culturally enriched interpretations.