Wiki-LLaVA: Hierarchical Retrieval-Augmented Generation for Multimodal LLMs

TL;DR

Wiki-LLaVA enhances multimodal large language models by integrating a hierarchical retrieval system to access external multimodal knowledge, significantly improving their ability to answer knowledge-dependent visual questions.

Contribution

The paper introduces Wiki-LLaVA, a novel hierarchical retrieval-augmented generation framework for multimodal LLMs, enabling effective external knowledge integration for visual question answering.

Findings

Improved accuracy in knowledge-dependent visual question answering

Effective retrieval of relevant multimodal passages from external sources

Enhanced dialogue precision with external knowledge augmentation

Abstract

Multimodal LLMs are the natural evolution of LLMs, and enlarge their capabilities so as to work beyond the pure textual modality. As research is being carried out to design novel architectures and vision-and-language adapters, in this paper we concentrate on endowing such models with the capability of answering questions that require external knowledge. Our approach, termed Wiki-LLaVA, aims at integrating an external knowledge source of multimodal documents, which is accessed through a hierarchical retrieval pipeline. Relevant passages, using this approach, are retrieved from the external knowledge source and employed as additional context for the LLM, augmenting the effectiveness and precision of generated dialogues. We conduct extensive experiments on datasets tailored for visual question answering with external data and demonstrate the appropriateness of our approach.