Generative Adversarial Perturbations

Omid Poursaeed, Isay Katsman, Bicheng Gao, Serge Belongie
2018 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition  
In this paper, we propose novel generative models for creating adversarial examples, slightly perturbed images resembling natural images but maliciously crafted to fool pre-trained models. We present trainable deep neural networks for transforming images to adversarial perturbations. Our proposed models can produce image-agnostic and image-dependent perturbations for targeted and nontargeted attacks. We also demonstrate that similar architectures can achieve impressive results in fooling both
more » ... assification and semantic segmentation models, obviating the need for hand-crafting attack methods for each task. Using extensive experiments on challenging high-resolution datasets such as ImageNet and Cityscapes, we show that our perturbations achieve high fooling rates with small perturbation norms. Moreover, our attacks are considerably faster than current iterative methods at inference time.
doi:10.1109/cvpr.2018.00465 dblp:conf/cvpr/PoursaeedKGB18 fatcat:gflwwbhp6fhe5pvbfthsezpfqa