Manual data augmentation pairing multiple Xs with one Y

I have 3 specific data augmentation and its hard for me to recreate that result in pytorch so I decided to create those augmentations using photoshop. Now I am trying to figure out what is the best practice of linking those premade data augmentation images with a single ground truth result. I am thinking of creating a csv file containing (Xn_1, Xn_2, Xn_3), (Yn) pairs. I would read the csv file and use it to load multiple random pairs in the data loader to make a single batch, eg X43_1, Y43 or X8_3, Y8. I have never done this way before so I was wondering if there a better way to approach this?

If they are named with a strict convention you could just code this convention in your dataset (e.g. my label-files are named just like the data files (except the file extension) and you could simply cut off the last number doing the path mapping.