1 paper
Shuntaro Okada, Kenji Doi, Ryota Yoshihashi +2
We propose a general framework for optimizing noise schedules in diffusion models, applicable to both training and sampling. Our method enforces a constant rate of change in the pr…