I would like to do an image classification task by Federated Learning. According to [tensorflow tutotial], 1, they download the original NIST dataset and use LEAF to pre-process them. However, in the tutorial, the details are omitted, they just load 'tff.simulation.datasets.emnist' to get eminist_train and emnist_test. Can anyone know the detail from LEAF's outputs (json files) to the accepted train/test datasets? Thanks