LFRAG: Layout-oriented Fine-grained Retrieval-Augmented Generation on Multimodal Document Understanding
The paper introduces LFRAG, a novel framework for multimodal document understanding that enhances retrieval-augmented generation. It shifts from page-level to block-level retrieval to improve accuracy and reduce redundancy in downstream tasks. Extensive experiments show LFRAG achieves state-of-the-art performance in retrieval tasks and significantly lowers token consumption in generation tasks.
- ▪LFRAG advances multimodal retrieval-augmented generation from page-level to block-level retrieval.
- ▪The framework utilizes layout segmentation to create semantically coherent fine-grained retrieval units.
- ▪LFRAG outperforms the best baseline by 7.20% in answer accuracy and reduces token consumption by 73.07%.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Information Retrieval arXiv:2605.22829 (cs) [Submitted on 18 Apr 2026] Title:LFRAG: Layout-oriented Fine-grained Retrieval-Augmented Generation on Multimodal Document Understanding Authors:Yifan Zhu, Yu Mi, Yue Lu, Yanchu Guan, Zhixuan Chu View a PDF of the paper titled LFRAG: Layout-oriented Fine-grained Retrieval-Augmented Generation on Multimodal Document Understanding, by Yifan Zhu and 4 other authors View PDF Abstract:Multimodal Retrieval-Augmented Generation (RAG) has emerged as an effective paradigm for enhancing Large Language Models (LLMs) with external knowledge.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
