Fine-grained Synthesis of Unrestricted Adversarial Examples

Omid Poursaeed, Tianxing Jiang, Yordanos Abraham Goshu, Harry Yang, Serge Belongie, Ser-Nam Lim

Keywords: adversarial examples, unrestricted attacks, generative models, adversarial training, generative adversarial networks

Abstract: We propose a novel approach for generating unrestricted adversarial examples by manipulating fine-grained aspects of image generation. Unlike existing unrestricted attacks that typically hand-craft geometric transformations, we learn stylistic and stochastic modifications leveraging state-of-the-art generative models. This allows us to manipulate an image in a controlled, fine-grained manner without being bounded by a norm threshold. Our approach can be used for targeted and non-targeted unrestricted attacks on classification, semantic segmentation and object detection models. Our attacks can bypass certified defenses, yet our adversarial images look indistinguishable from natural images as verified by human evaluation. Moreover, we demonstrate that adversarial training with our examples improves performance of the model on clean images without requiring any modifications to the architecture. We perform experiments on LSUN, CelebA-HQ and COCO-Stuff as high resolution datasets to validate efficacy of our proposed approach.

One-sentence Summary: A novel approach for fine-grained generation of unrestricted adversarial examples in classification, segmentation and detection which improves the model's performance on clean images.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Reviewed Version (pdf): https://openreview.net/references/pdf?id=NciuLABlQI