I am trying to scale up my model which uses a "cluster loss" extension, the implementation works so far on MNIST, but I would like to benefit from data augmentatio