You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Dec 14, 2023. It is now read-only.
But the idea is that the starting noise distribution does not have a mean of 0. For image-based diffusion models there is some evidence that this helps generate a more varied output distribution. It especially helps to generate very dark or very light images. See https://www.crosslabs.org/blog/diffusion-with-offset-noise for a more in depth discussion.
What is the effect of the offset noise in training?
The text was updated successfully, but these errors were encountered: