Dataloader for prepossessed images pytorch

I am doing some pre-processing algorithm on cifar-10 images which takes a lot of time. In order not to go through this process each time, It is preferred to save these images efficiently. It is about 10000 images, same as cifar-10.
I couldn’t find a coherent answer for that, how would you recommend to save these images in a separate dataloader?

You can save the pre-processed images onto the disk and load it using a separate Dataset instance