Scalable Diffusion Models with Transformers
Paper • 2212.09748 • Published • 17
复现经典的DiT工作(Scalable Diffusion Models with Transformers),训练数据为ImageNet.
代码仓库: https://github.com/lixiang90/ClassicalModels
vae.pt是用于图像压缩的vae模型,把(256,256,3)的图像压缩为(32,32,4)的latents.