Proximal Splitting Adversarial Attacks for Semantic Segmentation

TL;DR

This paper introduces a novel white-box adversarial attack tailored for semantic segmentation models, utilizing proximal splitting and augmented Lagrangian methods to generate smaller perturbations and outperform existing attacks.

Contribution

It presents the first effective adversarial attack for semantic segmentation using proximal splitting, addressing limitations of prior methods and providing a comprehensive benchmark.

Findings

Our attack produces significantly smaller $$ norms than previous methods.

It outperforms adapted classification attacks on segmentation tasks.

The approach handles numerous constraints efficiently within a nonconvex framework.

Abstract

Classification has been the focal point of research on adversarial attacks, but only a few works investigate methods suited to denser prediction tasks, such as semantic segmentation. The methods proposed in these works do not accurately solve the adversarial segmentation problem and, therefore, overestimate the size of the perturbations required to fool models. Here, we propose a white-box attack for these models based on a proximal splitting to produce adversarial perturbations with much smaller $\ell_\infty$ norms. Our attack can handle large numbers of constraints within a nonconvex minimization framework via an Augmented Lagrangian approach, coupled with adaptive constraint scaling and masking strategies. We demonstrate that our attack significantly outperforms previously proposed ones, as well as classification attacks that we adapted for segmentation, providing a first comprehensive benchmark for this dense task.