FlashEval: Towards Fast and Accurate Evaluation of Text-to-image Diffusion Generative Models