Don't Look into the Sun: Adversarial Solarization Attacks on Image Classifiers

TL;DR

This paper introduces a solarization-based adversarial attack on image classifiers that significantly degrades accuracy, revealing vulnerabilities in models even with some defenses and highlighting challenges in robustness evaluation.

Contribution

The paper presents a novel solarization attack method for image classifiers, offering a simple yet effective approach to evaluate robustness against out-of-distribution inputs.

Findings

The attack significantly reduces model accuracy on ImageNet.

Models trained with augmentations are not fully immune to the attack.

The attack can often be executed as a black-box with model-independent parameters.

Abstract

Assessing the robustness of deep neural networks against out-of-distribution inputs is crucial, especially in safety-critical domains like autonomous driving, but also in safety systems where malicious actors can digitally alter inputs to circumvent safety guards. However, designing effective out-of-distribution tests that encompass all possible scenarios while preserving accurate label information is a challenging task. Existing methodologies often entail a compromise between variety and constraint levels for attacks and sometimes even both. In a first step towards a more holistic robustness evaluation of image classification models, we introduce an attack method based on image solarization that is conceptually straightforward yet avoids jeopardizing the global structure of natural images independent of the intensity. Through comprehensive evaluations of multiple ImageNet models, we demonstrate the attack's capacity to degrade accuracy significantly, provided it is not integrated into the training augmentations. Interestingly, even then, no full immunity to accuracy deterioration is achieved. In other settings, the attack can often be simplified into a black-box attack with model-independent parameters. Defenses against other corruptions do not consistently extend to be effective against our specific attack. Project website: https://github.com/paulgavrikov/adversarial_solarization