Here we use Keras preprocessing layers to define the transformations for the training data (includes data augmentation),
and transformations for the validation data (only center cropping, resizing and normalizing).