Robust Unpaired Single Image Super-Resolution of Faces

TL;DR

This paper introduces a fast and effective adversarial attack method for facial class-specific single image super-resolution networks, improving the speed and effectiveness trade-off over existing methods.

Contribution

The paper presents a novel adversarial attack leveraging the MSE loss surface's parameterizable property to locate optimal degradations efficiently.

Findings

Achieves better speed vs effectiveness trade-off than FGSM and PGD.

Effective on unpaired facial and class-specific SISR.

Outperforms existing adversarial attacks in experiments.

Abstract

We propose an adversarial attack for facial class-specific Single Image Super-Resolution (SISR) methods. Existing attacks, such as the Fast Gradient Sign Method (FGSM) or the Projected Gradient Descent (PGD) method, are either fast but ineffective, or effective but prohibitively slow on these networks. By closely inspecting the surface that the MSE loss, used to train such networks, traces under varying degradations, we were able to identify its parameterizable property. We leverage this property to propose an adverasrial attack that is able to locate the optimum degradation (effective) without needing multiple gradient-ascent steps (fast). Our experiments show that the proposed method is able to achieve a better speed vs effectiveness trade-off than the state-of-theart adversarial attacks, such as FGSM and PGD, for the task of unpaired facial as well as class-specific SISR.