Resolution Dependent GAN Interpolation for Controllable Image Synthesis Between Domains

TL;DR

This paper introduces a resolution-dependent interpolation method between StyleGAN models, enabling controlled generation of images from novel domains with artistic direction.

Contribution

It proposes a novel interpolation technique that allows for domain transfer and control in image synthesis using StyleGAN models.

Findings

Enables generation of images from new domains

Provides controllability over generated imagery

Demonstrates effectiveness across different resolutions

Abstract

GANs can generate photo-realistic images from the domain of their training data. However, those wanting to use them for creative purposes often want to generate imagery from a truly novel domain, a task which GANs are inherently unable to do. It is also desirable to have a level of control so that there is a degree of artistic direction rather than purely curation of random results. Here we present a method for interpolating between generative models of the StyleGAN architecture in a resolution dependent manner. This allows us to generate images from an entirely novel domain and do this with a degree of control over the nature of the output.