Repo of the "Towards Out-of-Distribution Adversarial Robustness" paper https://arxiv.org/abs/2210.03150 .
WIP: Cleaning up/moving code from private repo here ASAP. Files are currently uploaded in portable format (i.e. .py files should be usable without the need for any other file provided requirements are installed). Note: Cloning this repo will downloaded the final MNIST and CIFAR10 models. The reason is that several of the results of the paper (namely, the attack discriminator, and transferability to CIFAR100 and SVHN) require those trained models as starting points to train other models. These models can be found in results/CIFAR10/models.