Research map / Image generation / Text-to-image diffusion
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
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use LLM-based instruction-following evaluation with multi-granularity visual descriptions rather than text-image matching | Text-to-image alignment evaluation · Concept Conjunction and Attribute Binding Contrast | Kendall's tau: 0.4058 -> 0.4871; 0.3768 -> 0.5167; 0.1982 -> 0.4005; 0.1500 -> 0.3955 | Improved |
| Use LLM-based instruction-following evaluation with multi-granularity visual descriptions rather than text-image matching | Text-to-image alignment evaluation · COCO2014, COCO2017, DrawBench, PaintSkills | Kendall's tau: 0.3465 -> 0.3629; 0.2793 -> 0.3357; 0.1569 -> 0.2230; 0.4911 -> 0.5600 | Improved |
| Use GPT-4 rather than GPT-3.5 as evaluator (ablation) | LLMScore · GeneralBench, Overall | Kendall's tau: 0.2480 -> 0.2793 | Improved |
| Use GPT-4 rather than GPT-3.5 as evaluator (ablation) | LLMScore · GeneralBench, Error Counting | Kendall's tau: 0.2205 -> 0.2131 | Hurt |
| Replace LLM reasoning with caption-to-prompt matching (ablation) | LLMScore evaluation pipeline · GeneralBench | Kendall's tau | Hurt |
| Use visual-description-to-prompt matching instead of LLM evaluation (ablation) | LLMScore evaluation pipeline · GeneralBench | Kendall's tau | Hurt |
| Use GPT-4 visual descriptions with sentence-matching metrics (ablation) | DescCLIP and DescMETEOR · GeneralBench | Kendall's tau: DescCLIP Overall: 0.1479 -> 0.1128; DescMETEOR Overall: 0.0042 -> 0.0297; DescCLIP Error Counting: 0.0467 -> 0.0149; Des | Mixed |
Closest papers by meaning
- Elucidating the design space of language models for image generation (ICML 2025 · similarity 0.89)
- SILMM: Self-Improving Large Multimodal Models for Compositional Text-to-Image Generation (CVPR 2025 · similarity 0.89)
- Revisiting text-to-image evaluation with Gecko: on metrics, prompts, and human rating (ICLR 2025 · similarity 0.88)
- A Comprehensive Study of Decoder-Only LLMs for Text-to-Image Generation (CVPR 2025 · similarity 0.88)
- Implicit Multimodal Alignment: On the Generalization of Frozen LLMs to Multimodal Inputs (NeurIPS 2024 · similarity 0.88)
- OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
