Image Synthesis with a Single (Robust) Classifier

TL;DR

This paper demonstrates that a single, adversarially robust classifier can be used for various challenging image synthesis tasks, highlighting the utility of robustness in machine learning.

Contribution

It introduces a minimal approach using a single robust classifier for multiple image synthesis tasks, contrasting with more complex state-of-the-art methods.

Findings

Robust classifiers can be directly used for image synthesis.

Adversarial robustness enables manipulation of salient input features.

The approach simplifies image synthesis workflows.

Abstract

We show that the basic classification framework alone can be used to tackle some of the most challenging tasks in image synthesis. In contrast to other state-of-the-art approaches, the toolkit we develop is rather minimal: it uses a single, off-the-shelf classifier for all these tasks. The crux of our approach is that we train this classifier to be adversarially robust. It turns out that adversarial robustness is precisely what we need to directly manipulate salient features of the input. Overall, our findings demonstrate the utility of robustness in the broader machine learning context. Code and models for our experiments can be found at https://git.io/robust-apps.