ExplainingAI

You are training on cpu as of now right ? Also can you confirm if your conda environment has python3.8 and have the requirements installed as mentioned in https://github.com/explainingai-code/DDPM-Pytorch/tree/main?tab=readme-ov-file#quickstart

Its because the images are 256x256 and by default the model config does downsampling only twice. Few things you can try to ensure you are able to train. 1. Resize...

Yes, I wanted to have this repo as a intro to diffusion which is why didn't want to add those and leave this as a bare minimum diffusion repo. I...

Hello @thatdev6 , this code expects the path to have png files. But seems like thats not the case for the path you have provided. Is it npy file? Cause...

Are you using the same code or have you made some modifications ? Your list at the end of dataset initialization is a list of numpy.ndarray objects(according to the error),...

The shapes of two images that your dataset returns are different (3x3264x2448 and 3x2448x3264).

I dont think the 3xwxh is an issue because the error says that your image shapes are 3xWxh so thats fine. But I think your path does not have all...

Yes center square crop of (2448x2448) and then resize to 64x64. How many images are there in your dataset?

Couple of things. Move the image reading to the data loader get_item method just like the code in repo. Simply collect the filenames from load_images method and nothing else. You...

Reduce the batch size to 16. That should work I think.