Read It Back: Pretrained MLLMs Are Zero-Shot Reward Models for Text-to-Image Generation

Benchmark Model Rank Results
text-to-image-generation-on-tiif-bench-miniSpectraReward (512px)#5Accuracy: 85.23
image-generation-on-wiseSelf-SpectraReward (w/ Self-CoT)#8Overall: 0.76
text-to-image-generation-on-dpg-benchSpectraReward (512px)#14DPG-Bench Score: 88.08
text-to-image-generation-on-genevalSelf-SpectraReward#27Overall: 0.898