Multi-Modal Language Models as Text-to-Image Model Evaluators
This addresses the need for efficient and human-aligned evaluation methods for T2I models, which is incremental as it builds on existing benchmarks but offers a novel approach.
The paper tackles the problem of evaluating text-to-image (T2I) generative models by proposing MT2IE, a framework that uses multi-modal large language models (MLLMs) as evaluator agents to assess prompt-generation consistency and image aesthetics, achieving the same relative model rankings as existing benchmarks with only 1/80th the number of prompts and showing higher correlation with human judgment.
The steady improvements of text-to-image (T2I) generative models lead to slow deprecation of automatic evaluation benchmarks that rely on static datasets, motivating researchers to seek alternative ways to evaluate the T2I progress. In this paper, we explore the potential of multi-modal large language models (MLLMs) as evaluator agents that interact with a T2I model, with the objective of assessing prompt-generation consistency and image aesthetics. We present Multimodal Text-to-Image Eval (MT2IE), an evaluation framework that iteratively generates prompts for evaluation, scores generated images and matches T2I evaluation of existing benchmarks with a fraction of the prompts used in existing static benchmarks. Moreover, we show that MT2IE's prompt-generation consistency scores have higher correlation with human judgment than scores previously introduced in the literature. MT2IE generates prompts that are efficient at probing T2I model performance, producing the same relative T2I model rankings as existing benchmarks while using only 1/80th the number of prompts for evaluation.