Adversarial Self-Defense for Cycle-Consistent GANs
This is the official implementation of the CycleGAN robust to self-adversarial attacks used in paper "Adversarial Self-Defense for Cycle-Consistent GANs" by Dina Bashkirova, Ben Usman and Kate Saenko, NeurIPS'2019. Paper
In this repository you can find the extension of the original implementation of CycleGAN in pytorch. This extension contains two defense techniques against the self-adversarial attacks performed by the unsupervised image-to-image translation methods that hold the cycle-consistency property. More information on self-adversarial attacks can be found in our paper.
In order to reproduce our results, you can find all the configuration files in configs directory.