Scholarlink

Large Vision-Language Models (LVLMs) have achieved remarkable performance in many vision-language tasks, yet their capabilities in fine-grained visual understanding remain insufficiently evaluated. Existing benchmarks either contain limited fine-grained evaluation tasks and/or samples or are confined to object-level assessments in natural images. To holistically assess LVLMs’ fine-grained visual understanding capabilities, we propose using document images with multi-granularity and multi-modal information to supplement natural images. In this light, we construct MMDocBench, a benchmark with various OCR-free document understanding tasks for the evaluation of fine-grained visual perception, reasoning and grounding abilities. MMDocBench defines 15 main tasks with 4, 338 QA pairs and 11, 353 supporting regions, covering various document images such as research papers, receipts, financial reports, Wikipedia tables, charts, and infographics. Based on MMDocBench, we conduct extensive experiments using 17 open-source and 5 proprietary advanced LVLMs, assessing their strengths and weaknesses across different tasks and document image types. The official website is https://mmdocbench.github.io/ .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MMDocBench: Benchmarking Large Vision-Language Models for Fine-Grained Visual Document Understanding and Grounding

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fengbin Zhu,
  • Ziyang Liu,
  • NG Xiang Yao,
  • Haohui Wu,
  • Wenjie Wang,
  • Fuli Feng,
  • Chao Wang,
  • Huanbo Luan,
  • Tat-Seng Chua

摘要

Large Vision-Language Models (LVLMs) have achieved remarkable performance in many vision-language tasks, yet their capabilities in fine-grained visual understanding remain insufficiently evaluated. Existing benchmarks either contain limited fine-grained evaluation tasks and/or samples or are confined to object-level assessments in natural images. To holistically assess LVLMs’ fine-grained visual understanding capabilities, we propose using document images with multi-granularity and multi-modal information to supplement natural images. In this light, we construct MMDocBench, a benchmark with various OCR-free document understanding tasks for the evaluation of fine-grained visual perception, reasoning and grounding abilities. MMDocBench defines 15 main tasks with 4, 338 QA pairs and 11, 353 supporting regions, covering various document images such as research papers, receipts, financial reports, Wikipedia tables, charts, and infographics. Based on MMDocBench, we conduct extensive experiments using 17 open-source and 5 proprietary advanced LVLMs, assessing their strengths and weaknesses across different tasks and document image types. The official website is https://mmdocbench.github.io/ .