Deep Image Destruction: Vulnerability of Deep Image-to-Image Models against Adversarial Attacks

TL;DR

This paper investigates the vulnerability of deep image-to-image models to adversarial attacks, revealing that their robustness varies significantly with attack methods and task objectives, and evaluates defense strategies.

Contribution

It provides the first comprehensive analysis of adversarial vulnerabilities in image-to-image models across multiple tasks and assesses the effectiveness of existing defenses.

Findings

Performance degradation varies by attack method and task.

Adversarial examples can transfer across different image-to-image tasks.

Conventional defenses have limited effectiveness against attacks.

Abstract

Recently, the vulnerability of deep image classification models to adversarial attacks has been investigated. However, such an issue has not been thoroughly studied for image-to-image tasks that take an input image and generate an output image (e.g., colorization, denoising, deblurring, etc.) This paper presents comprehensive investigations into the vulnerability of deep image-to-image models to adversarial attacks. For five popular image-to-image tasks, 16 deep models are analyzed from various standpoints such as output quality degradation due to attacks, transferability of adversarial examples across different tasks, and characteristics of perturbations. We show that unlike image classification tasks, the performance degradation on image-to-image tasks largely differs depending on various factors, e.g., attack methods and task objectives. In addition, we analyze the effectiveness of conventional defense methods used for classification models in improving the robustness of the image-to-image models.