Image Synthesis with a Single (Robust) Classifier
This work addresses the problem of simplifying image synthesis for researchers and practitioners by leveraging robustness, though it is incremental in applying existing robustness concepts to new tasks.
The paper tackles challenging image synthesis tasks by using a single adversarially robust classifier, enabling direct manipulation of input features without complex tools, achieving state-of-the-art results in tasks like style transfer and image inpainting.
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.