Benchmarking and Enhancing Text-to-Image Models for Generating Visual Representations in Early Arithmetic Education

2026-05-29

AI Summary

This work addresses the challenge that current text-to-image models struggle to accurately generate instructional visualizations that reflect the numerical values and structural relationships of arithmetic equations. It introduces, for the first time, the “equation-to-visual generation” task and constructs E2V-Bench, a benchmark comprising four pedagogically grounded visual formats derived from educational interviews and textbook analysis. The study also proposes automated metrics to evaluate the fidelity of object counts and relational structures in generated images. Building on this benchmark, a benchmark-guided model enhancement strategy is developed. Experimental results reveal significant deficiencies in mainstream models regarding counting accuracy and relational representation; while the proposed approach improves performance, it also exposes fundamental gaps in models’ understanding of numerical semantics and structural reasoning.

Abstract

AI systems are increasingly used to support educational content creation, yet it remains unclear whether they can generate outputs that faithfully represent the pedagogical concepts they are intended to teach. Thus, we introduce equation-to-visual generation, a task that, in contrast to conventional image generation, requires producing pedagogically meaningful visuals from arithmetic equations while precisely preserving their numerical and relational structure. Informed by interviews with teachers and an analysis of educational materials, we construct E2V-Bench, a benchmark spanning four pedagogically grounded visual types, along with automatic metrics for evaluating visual correctness. Our evaluation reveals that recent text-to-image (T2I) models frequently fail on this task, with errors dominated by incorrect object counts and broken relational structure. Building on this, we explore benchmark-guided enhancement strategies. These strategies improve representative models, while the remaining gap calls for stronger numerical and relational grounding in future T2I models.

Problem

• text-to-image generation
• visual representation
• arithmetic education
• numerical grounding
• relational structure

Innovation

• equation-to-visual generation
• educational AI
• text-to-image models
• numerical grounding
• relational structure