LLMScore: Unveiling the Power of Large Language Models in Text-to-Image Synthesis Evaluation

Yujie Lu, Xianjun Yang, Xiujun Li, Xin Eric Wang, William Yang Wang

Keywords: Text-to-Image Evaluation, Visio-linguistic Compositionality, Large Language Models

TL;DR: We propose LLMScore, a new framework that can offer interpretable evaluation scores to measure how well the generated images are aligned with text prompts for text-to-image evaluation.

Abstract:

Existing automatic evaluation on text-to-image synthesis can only provide an image-text matching score, without considering the object-level compositionality, which results in poor correlation with human judgments. In this work, we propose LLMScore, a new framework that offers evaluation scores with multi-granularity compositionality. LLMScore leverages the large language models (LLMs) to evaluate text-to-image models. Initially, it transforms the image into image-level and object-level visual descriptions. Then an evaluation instruction is fed into the LLMs to measure the alignment between the synthesized image and the text, ultimately generating a score accompanied by a rationale.
Our substantial analysis reveals the highest correlation of LLMScore with human judgments on a wide range of datasets (Attribute Binding Contrast, Concept Conjunction, MSCOCO, DrawBench, PaintSkills). Notably, our LLMScore achieves Kendall's tau correlation with human evaluations that is 58.8% and 31.2% higher than the commonly-used text-image matching metrics CLIP and BLIP, respectively.

Submission Number: 4416