How to create dynamic Pytorch Dataloader

Question

I’m doing active learning for my segmentation project. I have an “IMG folder” containing about 5000 images. My task is: First, loading 500 images of that IMG folder for training