LLMScore: Unveiling the Power of Large Language Models in Text-to-Image Synthesis Evaluation
This addresses the need for better automatic evaluation in text-to-image synthesis, offering a more human-aligned metric for researchers and practitioners, though it is incremental as it builds on existing LLM capabilities.
The paper tackles the problem of evaluating text-to-image synthesis by proposing LLMScore, a framework that uses large language models to provide multi-granularity compositionality scores, achieving up to 58.8% higher correlation with human judgments than existing metrics like CLIP.
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.