Search papers, labs, and topics across Lattice.
This paper introduces CURV, a curriculum learning framework designed to enhance chart question answering (CQA) by reformulating it into multi-step visual grounded reasoning, effectively integrating logical reasoning with visual comprehension. By employing a three-level curriculum dataset (CCQA) that scales synthetic generation across various chart types and reasoning patterns, CURV systematically trains models from basic to complex reasoning tasks. Experimental results show CURV achieves significant performance improvements, with up to 20.50% gains over baseline models and strong generalization to real-world benchmarks and out-of-domain tasks.
CURV transforms chart question answering by embedding dynamic visual grounding into a structured learning curriculum, leading to unprecedented improvements in reasoning accuracy.
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/.
