MV-MATH: Evaluating Multimodal Math Reasoning in Multi-Visual Contexts

TL;DR

MV-MATH introduces a new benchmark dataset with multi-visual math problems to evaluate multimodal models' reasoning abilities in realistic, complex visual-text scenarios, revealing significant challenges faced by current models.

Contribution

We created MV-MATH, a comprehensive dataset of multi-visual math problems from real-world scenarios, and evaluated MLLMs' performance, highlighting existing gaps and challenges.

Findings

MLLMs perform substantially worse than humans on MV-MATH.

Current models struggle with multi-visual reasoning tasks.

Error analysis reveals specific reasoning challenges.

Abstract

Multimodal Large Language Models (MLLMs) have shown promising capabilities in mathematical reasoning within visual contexts across various datasets. However, most existing multimodal math benchmarks are limited to single-visual contexts, which diverges from the multi-visual scenarios commonly encountered in real-world mathematical applications. To address this gap, we introduce MV-MATH: a meticulously curated dataset of 2,009 high-quality mathematical problems. Each problem integrates multiple images interleaved with text, derived from authentic K-12 scenarios, and enriched with detailed annotations. MV-MATH includes multiple-choice, free-form, and multi-step questions, covering 11 subject areas across 3 difficulty levels, and serves as a comprehensive and rigorous benchmark for assessing MLLMs' mathematical reasoning in multi-visual contexts. Through extensive experimentation, we observe that MLLMs encounter substantial challenges in multi-visual math tasks, with a considerable performance gap relative to human capabilities on MV-MATH. Furthermore, we analyze the performance and error patterns of various models, providing insights into MLLMs' mathematical reasoning capabilities within multi-visual settings.