How to Utilize Complementary Vision-Text Information for 2D Structure Understanding
This work addresses the challenge of accurately interpreting 2D tables for applications in data processing and AI, representing an incremental improvement over existing multimodal methods.
The paper tackles the problem of 2D structure understanding by addressing the limitations of LLMs and visual encoders in handling tables, proposing DiVA-Former to integrate vision and text information, which improves upon a pure-text baseline by 23.9% across 13 benchmarks.
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.