The shape and simplicity biases of adversarially robust imagenet-trained cnns

Publication
arXiv preprint arXiv:2006.09373