Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Decoupling Skeleton and Flesh: Efficient Multimodal Table Reasoning with Disentangled Alignment and Structure-aware Guidance

Yingjie Zhu, Xuefeng Bai, Kehai Chen, Yang Xiang, Youcheng Pan, Xiaoqiang Zhou, Min Zhang

TL;DR

This work tackles multimodal table reasoning with LVLMs by addressing the entanglement of table structure and content. It introduces DiSCo to disentangle structure and semantics during alignment and Table-GLS to perform three-stage, training-free reasoning guided by global-to-local structure signals. Empirical results across 21 benchmarks show strong improvements in both table understanding and reasoning, with notable generalization to unseen table structures and reduced reliance on large-scale annotations or external tools. The combined framework enables more robust, interpretable reasoning over complex tables in real-world settings, with practical implications for scalable multimodal analytics.

Abstract

Reasoning over table images remains challenging for Large Vision-Language Models (LVLMs) due to complex layouts and tightly coupled structure-content information. Existing solutions often depend on expensive supervised training, reinforcement learning, or external tools, limiting efficiency and scalability. This work addresses a key question: how to adapt LVLMs to table reasoning with minimal annotation and no external tools? Specifically, we first introduce DiSCo, a Disentangled Structure-Content alignment framework that explicitly separates structural abstraction from semantic grounding during multimodal alignment, efficiently adapting LVLMs to tables structures. Building on DiSCo, we further present Table-GLS, a Global-to-Local Structure-guided reasoning framework that performs table reasoning via structured exploration and evidence-grounded inference. Extensive experiments across diverse benchmarks demonstrate that our framework efficiently enhances LVLM's table understanding and reasoning capabilities, particularly generalizing to unseen table structures.

Decoupling Skeleton and Flesh: Efficient Multimodal Table Reasoning with Disentangled Alignment and Structure-aware Guidance

TL;DR

This work tackles multimodal table reasoning with LVLMs by addressing the entanglement of table structure and content. It introduces DiSCo to disentangle structure and semantics during alignment and Table-GLS to perform three-stage, training-free reasoning guided by global-to-local structure signals. Empirical results across 21 benchmarks show strong improvements in both table understanding and reasoning, with notable generalization to unseen table structures and reduced reliance on large-scale annotations or external tools. The combined framework enables more robust, interpretable reasoning over complex tables in real-world settings, with practical implications for scalable multimodal analytics.

Abstract

Reasoning over table images remains challenging for Large Vision-Language Models (LVLMs) due to complex layouts and tightly coupled structure-content information. Existing solutions often depend on expensive supervised training, reinforcement learning, or external tools, limiting efficiency and scalability. This work addresses a key question: how to adapt LVLMs to table reasoning with minimal annotation and no external tools? Specifically, we first introduce DiSCo, a Disentangled Structure-Content alignment framework that explicitly separates structural abstraction from semantic grounding during multimodal alignment, efficiently adapting LVLMs to tables structures. Building on DiSCo, we further present Table-GLS, a Global-to-Local Structure-guided reasoning framework that performs table reasoning via structured exploration and evidence-grounded inference. Extensive experiments across diverse benchmarks demonstrate that our framework efficiently enhances LVLM's table understanding and reasoning capabilities, particularly generalizing to unseen table structures.
Paper Structure (34 sections, 7 equations, 6 figures, 11 tables)

This paper contains 34 sections, 7 equations, 6 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Table Modeling
  4. Multimodal Table Understanding and Reasoning
  5. Methodology
  6. Disentangled Structure–Content Alignment
  7. Structure Alignment.
  8. Content Alignment.
  9. Global-to-Local Structure-Guided Reasoning
  10. Global Structure Exploration.
  11. Self-refined Sub-table Extraction.
  12. Evidence-grounded Reasoning.
  13. Experimental Setup
  14. Datasets
  15. Baselines
  16. ...and 19 more sections

Figures (6)

  • Figure 1: Comparison of our framework with current methods.
  • Figure 2: Overall framework of DiSCo and Table-GLS.
  • Figure 3: Model performance on representative understanding tasks across various table layouts.
  • Figure 4: Correlation between table understanding and reasoning performance of Qwen3-VL-DiSCo.
  • Figure 5: An example of multimodal table reasoning task.
  • ...and 1 more figures