DocLens : A Tool-Augmented Multi-Agent Framework for Long Visual Document Understanding
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914158724775936 |
|---|---|
| author | Zhu, Dawei Meng, Rui Chen, Jiefeng Li, Sujian Pfister, Tomas Yoon, Jinsung |
| author_facet | Zhu, Dawei Meng, Rui Chen, Jiefeng Li, Sujian Pfister, Tomas Yoon, Jinsung |
| contents | Comprehending long visual documents, where information is distributed across extensive pages of text and visual elements, is a critical but challenging task for modern Vision-Language Models (VLMs). Existing approaches falter on a fundamental challenge: evidence localization. They struggle to retrieve relevant pages and overlook fine-grained details within visual elements, leading to limited performance and model hallucination. To address this, we propose DocLens, a tool-augmented multi-agent framework that effectively ``zooms in'' on evidence like a lens. It first navigates from the full document to specific visual elements on relevant pages, then employs a sampling-adjudication mechanism to generate a single, reliable answer. Paired with Gemini-2.5-Pro, DocLens achieves state-of-the-art performance on MMLongBench-Doc and FinRAGBench-V, surpassing even human experts. The framework's superiority is particularly evident on vision-centric and unanswerable queries, demonstrating the power of its enhanced localization capabilities. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_11552 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | DocLens : A Tool-Augmented Multi-Agent Framework for Long Visual Document Understanding Zhu, Dawei Meng, Rui Chen, Jiefeng Li, Sujian Pfister, Tomas Yoon, Jinsung Computer Vision and Pattern Recognition Computation and Language Comprehending long visual documents, where information is distributed across extensive pages of text and visual elements, is a critical but challenging task for modern Vision-Language Models (VLMs). Existing approaches falter on a fundamental challenge: evidence localization. They struggle to retrieve relevant pages and overlook fine-grained details within visual elements, leading to limited performance and model hallucination. To address this, we propose DocLens, a tool-augmented multi-agent framework that effectively ``zooms in'' on evidence like a lens. It first navigates from the full document to specific visual elements on relevant pages, then employs a sampling-adjudication mechanism to generate a single, reliable answer. Paired with Gemini-2.5-Pro, DocLens achieves state-of-the-art performance on MMLongBench-Doc and FinRAGBench-V, surpassing even human experts. The framework's superiority is particularly evident on vision-centric and unanswerable queries, demonstrating the power of its enhanced localization capabilities. |
| title | DocLens : A Tool-Augmented Multi-Agent Framework for Long Visual Document Understanding |
| topic | Computer Vision and Pattern Recognition Computation and Language |
| url | https://arxiv.org/abs/2511.11552 |