Evaluating Numerical Reasoning in Text-to-Image Models
This work identifies a critical limitation in text-to-image models for applications requiring precise numerical understanding, such as educational or technical content generation.
The paper evaluated text-to-image models on numerical reasoning tasks and found that even advanced models have limited numerical skills, such as generating exact object counts only for small numbers and struggling with linguistic quantifiers and fractional concepts.
Text-to-image generative models are capable of producing high-quality images that often faithfully depict concepts described using natural language. In this work, we comprehensively evaluate a range of text-to-image models on numerical reasoning tasks of varying difficulty, and show that even the most advanced models have only rudimentary numerical skills. Specifically, their ability to correctly generate an exact number of objects in an image is limited to small numbers, it is highly dependent on the context the number term appears in, and it deteriorates quickly with each successive number. We also demonstrate that models have poor understanding of linguistic quantifiers (such as "a few" or "as many as"), the concept of zero, and struggle with more advanced concepts such as partial quantities and fractional representations. We bundle prompts, generated images and human annotations into GeckoNum, a novel benchmark for evaluation of numerical reasoning.