文本生成图像研究综述

曹寅,秦俊平,马千里,孙昊,闫凯,王磊,任家琪

Survey of text-to-image synthesis

Yin CAO,Junping QIN,Qianli MA,Hao SUN,Kai YAN,Lei WANG,Jiaqi REN

表 2 基于自回归模型架构和扩散模型架构的文本生成图像方法对比

Tab.2 Comparison of text-to-image generation methods based on autoregressive model architecture and diffusion model architecture

方法 MS-COCO数据集 FID Zero-shot FID DALL-E[15] — 28.0 ERNIE-ViLG[73] — 14.7 CogView [16] — 27.1 CogView2[17] 17.7 24.0 Parti[18] 3.22 7.23 KNN-Diffusion[75] 16.66 — GLIDE[76] — 12.24 DALL-E 2[77] — 10.39 Imagen[78] — 7.27 Stable Diffusion[79] — 12.63 Re-Imagen[19] 5.25 6.88