Using improved-diffusion code for 3D data

[40uf411]

Hi,

I am trying your code to train on 3D data but so far I am having issues with which parameters to change to make it work. A sample from my data is 50x150x150, and so far, I have changed the the dim in the UNET and added this to the create model function: elif image_size == 150 channel_mult = (1, 2, 2, 4, 4) But I am still having some errors due to Unet architecture.

Is there any guide on how to port this to 3D?

Thanks.