Paper
How to Utilize Complementary Vision-Text Information for 2D Structure Understanding
Authors
Jiancheng Dong, Pengyue Jia, Derong Xu, Jiawei Cheng, Jingyu Peng, Chao Zhang, Bowen Liu, Xin Sun, Lixin Su, Shuaiqiang Wang, Dawei Yin, Xiangyu Zhao
Abstract
LLMs typically linearize 2D tables into 1D sequences to fit their autoregressive architecture, which weakens row-column adjacency and other layout cues. In contrast, purely visual encoders can capture spatial cues, yet often struggle to preserve exact cell text. Our analysis reveals that these two modalities provide highly distinct information to LLMs and exhibit strong complementarity. However, direct concatenation and other fusion methods yield limited gains and frequently introduce cross-modal interference. To address this issue, we propose DiVA-Former, a lightweight architecture designed to effectively integrate vision and text information. DiVA-Former leverages visual tokens as dynamic queries to distill long textual sequences into digest vectors, thereby effectively exploiting complementary vision--text information. Evaluated across 13 table benchmarks, DiVA-Former improves upon the pure-text baseline by 23.9\% and achieves consistent gains over existing baselines using visual inputs, textual inputs, or a combination of both.
Metadata
Related papers
Vibe Coding XR: Accelerating AI + XR Prototyping with XR Blocks and Gemini
Ruofei Du, Benjamin Hersh, David Li, Nels Numan, Xun Qian, Yanhe Chen, Zhongy... • 2026-03-25
Comparing Developer and LLM Biases in Code Evaluation
Aditya Mittal, Ryan Shar, Zichu Wu, Shyam Agarwal, Tongshuang Wu, Chris Donah... • 2026-03-25
The Stochastic Gap: A Markovian Framework for Pre-Deployment Reliability and Oversight-Cost Auditing in Agentic Artificial Intelligence
Biplab Pal, Santanu Bhattacharya • 2026-03-25
Retrieval Improvements Do Not Guarantee Better Answers: A Study of RAG for AI Policy QA
Saahil Mathur, Ryan David Rittner, Vedant Ajit Thakur, Daniel Stuart Schiff, ... • 2026-03-25
MARCH: Multi-Agent Reinforced Self-Check for LLM Hallucination
Zhuo Li, Yupeng Zhang, Pengyu Cheng, Jiajun Song, Mengyu Zhou, Hao Li, Shujie... • 2026-03-25
Raw Data (Debug)
{
"raw_xml": "<entry>\n <id>http://arxiv.org/abs/2603.16245v1</id>\n <title>How to Utilize Complementary Vision-Text Information for 2D Structure Understanding</title>\n <updated>2026-03-17T08:30:01Z</updated>\n <link href='https://arxiv.org/abs/2603.16245v1' rel='alternate' type='text/html'/>\n <link href='https://arxiv.org/pdf/2603.16245v1' rel='related' title='pdf' type='application/pdf'/>\n <summary>LLMs typically linearize 2D tables into 1D sequences to fit their autoregressive architecture, which weakens row-column adjacency and other layout cues. In contrast, purely visual encoders can capture spatial cues, yet often struggle to preserve exact cell text. Our analysis reveals that these two modalities provide highly distinct information to LLMs and exhibit strong complementarity. However, direct concatenation and other fusion methods yield limited gains and frequently introduce cross-modal interference. To address this issue, we propose DiVA-Former, a lightweight architecture designed to effectively integrate vision and text information. DiVA-Former leverages visual tokens as dynamic queries to distill long textual sequences into digest vectors, thereby effectively exploiting complementary vision--text information. Evaluated across 13 table benchmarks, DiVA-Former improves upon the pure-text baseline by 23.9\\% and achieves consistent gains over existing baselines using visual inputs, textual inputs, or a combination of both.</summary>\n <category scheme='http://arxiv.org/schemas/atom' term='cs.CV'/>\n <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n <published>2026-03-17T08:30:01Z</published>\n <arxiv:comment>16 pages, 5 figures</arxiv:comment>\n <arxiv:primary_category term='cs.CV'/>\n <author>\n <name>Jiancheng Dong</name>\n </author>\n <author>\n <name>Pengyue Jia</name>\n </author>\n <author>\n <name>Derong Xu</name>\n </author>\n <author>\n <name>Jiawei Cheng</name>\n </author>\n <author>\n <name>Jingyu Peng</name>\n </author>\n <author>\n <name>Chao Zhang</name>\n </author>\n <author>\n <name>Bowen Liu</name>\n </author>\n <author>\n <name>Xin Sun</name>\n </author>\n <author>\n <name>Lixin Su</name>\n </author>\n <author>\n <name>Shuaiqiang Wang</name>\n </author>\n <author>\n <name>Dawei Yin</name>\n </author>\n <author>\n <name>Xiangyu Zhao</name>\n </author>\n </entry>"
}