Saved in:
Bibliographic Details
Main Authors: Chen, Jian, Li, Ming, Kil, Jihyung, Wang, Chenguang, Yu, Tong, Rossi, Ryan, Zhou, Tianyi, Chen, Changyou, Zhang, Ruiyi
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.07493
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908499914522624
author Chen, Jian
Li, Ming
Kil, Jihyung
Wang, Chenguang
Yu, Tong
Rossi, Ryan
Zhou, Tianyi
Chen, Changyou
Zhang, Ruiyi
author_facet Chen, Jian
Li, Ming
Kil, Jihyung
Wang, Chenguang
Yu, Tong
Rossi, Ryan
Zhou, Tianyi
Chen, Changyou
Zhang, Ruiyi
contents Most organizational data in this world are stored as documents, and visual retrieval plays a crucial role in unlocking the collective intelligence from all these documents. However, existing benchmarks focus on English-only document retrieval or only consider multilingual question-answering on a single-page image. To bridge this gap, we introduce VisR-Bench, a multilingual benchmark designed for question-driven multimodal retrieval in long documents. Our benchmark comprises over 35K high-quality QA pairs across 1.2K documents, enabling fine-grained evaluation of multimodal retrieval. VisR-Bench spans sixteen languages with three question types (figures, text, and tables), offering diverse linguistic and question coverage. Unlike prior datasets, we include queries without explicit answers, preventing models from relying on superficial keyword matching. We evaluate various retrieval models, including text-based methods, multimodal encoders, and MLLMs, providing insights into their strengths and limitations. Our results show that while MLLMs significantly outperform text-based and multimodal encoder models, they still struggle with structured tables and low-resource languages, highlighting key challenges in multilingual visual retrieval.
format Preprint
id arxiv_https___arxiv_org_abs_2508_07493
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle VisR-Bench: An Empirical Study on Visual Retrieval-Augmented Generation for Multilingual Long Document Understanding
Chen, Jian
Li, Ming
Kil, Jihyung
Wang, Chenguang
Yu, Tong
Rossi, Ryan
Zhou, Tianyi
Chen, Changyou
Zhang, Ruiyi
Computer Vision and Pattern Recognition
Most organizational data in this world are stored as documents, and visual retrieval plays a crucial role in unlocking the collective intelligence from all these documents. However, existing benchmarks focus on English-only document retrieval or only consider multilingual question-answering on a single-page image. To bridge this gap, we introduce VisR-Bench, a multilingual benchmark designed for question-driven multimodal retrieval in long documents. Our benchmark comprises over 35K high-quality QA pairs across 1.2K documents, enabling fine-grained evaluation of multimodal retrieval. VisR-Bench spans sixteen languages with three question types (figures, text, and tables), offering diverse linguistic and question coverage. Unlike prior datasets, we include queries without explicit answers, preventing models from relying on superficial keyword matching. We evaluate various retrieval models, including text-based methods, multimodal encoders, and MLLMs, providing insights into their strengths and limitations. Our results show that while MLLMs significantly outperform text-based and multimodal encoder models, they still struggle with structured tables and low-resource languages, highlighting key challenges in multilingual visual retrieval.
title VisR-Bench: An Empirical Study on Visual Retrieval-Augmented Generation for Multilingual Long Document Understanding
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.07493