Re-Implementation of "mixup: BEYOND EMPIRICAL RISK MINIMIZATION" experimental setups (vgg16 with batch norm) epochs = 100 batch_size = 100 optimizer = SGD results mixup normal accuracy 94.02 [%] 93.52 [%] Usage of example (additional setup is written in configure.py or train.py) python3 train.py --alpha 0.2