Adversarially robust segmentation models learn perceptually-aligned gradients

Pedro Sandoval-Segura

The effects of adversarial training on semantic segmentation networks has not been thoroughly explored. While previous work has shown that adversarially-trained image classifiers c…