CURV: Enhancing Chart Understanding Through Curriculum Visual Grounded Reasoning

2026-08-03

AI Summary

Existing multimodal large language models lack visual grounding reasoning capabilities in chart question answering, leading to perceptual biases and a disconnect between perception and reasoning. This work proposes CURV, a curriculum learning framework that reformulates the task as multi-step visual grounding reasoning, where each step integrates logical inference with dynamic visual grounding guided by spatial attention. To support progressive training, the authors construct CCQA, a three-tier synthetic dataset. CURV is the first approach to internalize a dynamic visual reasoning mechanism, achieving substantial performance gains: up to 20.50% improvement on standard benchmarks, along with 12.30% and 10.20% gains on real-world chart question answering and cross-domain multimodal reasoning tasks, respectively.

Abstract

Chart question answering (CQA) requires multimodal large language models (MLLMs) to integrate visual comprehension with logical reasoning, yet current models struggle with accurate visual grounding and coherent reasoning chains. While extrinsic chain-of-thought prompting and visual cues significantly improve performance, current MLLMs lack intrinsic visual grounded reasoning capabilities, leading to inaccurate perception and reasoning disconnected from visual evidence. To address these limitations, we propose CURV, a curriculum learning framework that develops intrinsic visual reasoning capabilities by reformulating CQA as multi-step visual grounded reasoning, where each step coordinates logical reasoning with dynamic visual grounding through spatial attention concentration. To assist model learning, we further introduce CCQA, a three-level curriculum dataset with scalable synthetic generation across diverse chart types and reasoning patterns. Our curriculum systematically progresses from basic single-operation reasoning to complex multi-chart compositional tasks. Experiments demonstrate that CURV achieves up to $\uparrow20.50\%$ improvements over baselines and is generalizable to real-world benchmarks (up to $\uparrow12.30\%$) and out-of-domain multimodal reasoning tasks (up to $\uparrow10.20\%$), validating the effectiveness of internalizing visual reasoning with dynamic grounding for enhanced chart understanding capabilities. Code is available at: https://xhguo7.github.io/CURV/.

Problem

• chart question answering
• visual grounded reasoning
• multimodal large language models
• visual grounding
• reasoning chains

Innovation

• curriculum learning
• visual grounded reasoning
• multimodal large language models
• chart question answering
• dynamic visual grounding