ExplainingAI

Hello, 224x224 is still large for this model. Can you please try to follow the steps mentioned [here](https://github.com/explainingai-code/DDPM-Pytorch/issues/1#issuecomment-1862244458) and see if it works fine after that ?

With 224x224 images, using the current code version it would be difficult, but you could try the following: 1. Reduce the number of channels and layers significantly until single gpu...

When you say results are not as expected, do you mean images generated are completely garbage or they are just not of that high quality ? Was the generation output...

Couple of things that I can think of. I see your images are grayscale, any specific reason to use 3 channels. Maybe try with im_channels : 1 Based on these...

Hi @xiaoxiao079 , It looks from the error that code is trying to load a checkpoint which is trained on a different than what you are currently using to train/infer....

Hi, This DDPM code was created replicating the blocks from the official version for stable diffusion with the goal to create a minimalistic diffusion model purely for understanding purposes(and then...

Hi @vinayak-sharan , Could you tell me which dataset are you training it on and what is the conditioning(text/mask/class) that you are trying ?

I dont have any logs but I think for this case(CelebHQ conditioned on masks and texts), by 50 epochs you should get decent generation output. By any chance have you...

I was never able to train for more than 100 epochs(cause of compute limitations), but the issue of increase in loss, I think should be reduced by adding a decay...

Hello @vinodrajendran001, I think based on your requirement you can use the class embeddings(for categorical) and timestep embeddings for numerical conditioning. Lets go through the categorical conditions first. Assume you...