Latent Space Conditioning on Generative Adversarial Networks

TL;DR

This paper introduces a novel framework for unsupervised conditional GANs that leverages latent space structure to control image generation without relying on labeled data, maintaining high sample quality.

Contribution

It proposes a new method combining adversarial and representation learning to condition GANs using latent space features instead of labels, reducing dependence on annotated data.

Findings

Produces high-quality samples on demand

Breaks dependency between condition and label

Operates effectively in an unsupervised setting

Abstract

Generative adversarial networks are the state of the art approach towards learned synthetic image generation. Although early successes were mostly unsupervised, bit by bit, this trend has been superseded by approaches based on labelled data. These supervised methods allow a much finer-grained control of the output image, offering more flexibility and stability. Nevertheless, the main drawback of such models is the necessity of annotated data. In this work, we introduce an novel framework that benefits from two popular learning techniques, adversarial training and representation learning, and takes a step towards unsupervised conditional GANs. In particular, our approach exploits the structure of a latent space (learned by the representation learning) and employs it to condition the generative model. In this way, we break the traditional dependency between condition and label, substituting the latter by unsupervised features coming from the latent space. Finally, we show that this new technique is able to produce samples on demand keeping the quality of its supervised counterpart.