How to train on own dataset that already has train, val and test subfolders?

My dataset has three subfolders - train, val and test. Each of these subfolders have two classes - class1 and class2. How do I use the classification pipeline script to train for this particular setup?

The current script splits the entire data into 5 folds (with 1 val fold) on the fly. But, I want to train for this already existing dataset splitup.