Report
iDDM is claimed to deliver few-step diffusion without a teacher
A post says iDDM trains a single model from scratch in one stage, without self-distillation or JVPs.
TLDR
A post describes iDDM as a single diffusion model trained from scratch in one stage, without a teacher, self-distillation or JVPs. It reports ImageNet-256 results of 4.48 FID at four steps and 2.38 at 50 steps from the same checkpoint.
Combined views
3
1 Source, first seen ago
