CVPR2023
Dimensionality-Varying Diffusion Process
Han Zhang, Ruili Feng, Zhantao Yang, Lianghua Huang, Yu Liu, Yifei Zhang, Yujun Shen, Deli Zhao, Jingren Zhou, Fan Cheng
摘要
Figure 1. Synthesized samples on various datasets, including FFHQ (1024 2 and 256 2 ), LSUN Church (256 2 ), LSUN Bedroom (256 2 ), LSUN Cat ( 256 2 ) and CIFAR10 (32 2 ). All these samples are generated from a 64 2 noise except CIFAR10 from 16 2 , while conventional diffusion models can only start from a noise with the same dimension as the final sample.