README Augment training images for deep neural networks by randomizing their visual style, as described in our paper: https://arxiv.org/abs/1809.05375