High magnitude output

[aziere-vivodyne]

Hello,

First off, thank you so much for this great work.
I am trying to train it on a custom dataset of z-stack similar to what is in OpenSRH but with different channels.
I am running into an issue with the model where the generated pixel values vary between ~(-1000,1000) after going through the DDIM sampler model.

I use the exact same config than in train_msdsr.yaml for the model parameters.

Any insight on why it is the case?