Quality Assessment for Text-to-Image Generation: A Survey

In recent years, there have been notable advancements in text-to-image generation facilitated by artificial intelligence (AI) technology. Text-to-image generation requires higher-level cognitive abilities, posing unique challenges for image quality assessment typically designed for professionally generated content and user-generated content. Existing works have extensively investigated quality assessment from subjective and objective perspectives, covering a range of evaluation dimensions such as text–image alignment, perception, esthetics, fairness, and toxicity. This article provides a comprehensive overview of recent advancements in image quality assessment for text-to-image generation. In particular, we review existing quality assessment studies from subjective and objective perspectives, highlighting representative datasets and objective metrics for assessing different aspects of AI-generated image quality. Additionally, we discuss the limitations of current research and propose future directions.