MathVis-Fine: Aligning Visual Supervision with Necessity via Progressive Dependency-Guided Training for Multimodal Mathematical Reasoning

2026-06-16

AI Summary

Existing multimodal mathematical reasoning approaches treat visual inputs as homogeneous or auxiliary signals, overlooking instance-level image-text dependencies, which results in coarse-grained visual supervision and misaligned training feedback. To address this limitation, this work introduces MathVis-Fine, the first dataset annotated with fine-grained visual dependency labels, and proposes a two-stage progressive visual enhancement training paradigm. By incorporating an instance-level visual dependency assessment mechanism, the framework dynamically balances answer correctness with visual grounding rewards, enabling on-demand alignment of visual supervision signals. Experimental results demonstrate that the proposed approach adaptively enhances model perception capabilities according to the degree of visual dependency, significantly improving performance on multimodal mathematical reasoning tasks.

Abstract

Chain-of-Thought (CoT) reasoning has extended from purely linguistic domains to multimodal scenarios; however, existing approaches often treat visual inputs as homogeneous or auxiliary signals, failing to capture the intricate and sample-specific dependencies between text and images in mathematical problem-solving. This gives rise to two core issues: first, the supervisory signals for visual content are generalized and coarse-grained, lacking adaptation to the actual necessity of visual information in each sample; second, training feedback becomes inaccurate when visual rewards are uniformly applied without distinguishing the complementary relationships among inputs. These limitations hinder models from achieving precise multimodal reasoning. In this work, we propose a framework for modeling fine-grained visual dependencies in mathematical reasoning. We first construct the MathVis-Fine dataset, augmenting fine-grained visual annotations with visual dependency ratings. Building upon this dataset, we introduce a two-stage progressive visual enhancement training paradigm that balances answer correctness rewards and visual grounding rewards according to the intrinsic visual dependency level of each sample, thereby mitigating reward bias and improving supervision accuracy. Extensive experiments demonstrate that the MathVis-Fine framework effectively enhances visual perception progressively based on visual dependency, offering a more precise training framework for multimodal mathematical reasoning. We will release the dataset upon acceptance.

Problem

• multimodal mathematical reasoning
• visual supervision
• visual dependency
• reward bias
• fine-grained annotation

Innovation

• visual dependency
• progressive training
• multimodal mathematical reasoning
• fine-grained supervision
• reward balancing