Unsupervised Projection Networks for Generative Adversarial Networks

TL;DR

This paper introduces an unsupervised projection network that maps images onto the latent space of a trained GAN, enabling tasks like super-resolution and semantic clustering without additional supervision.

Contribution

It presents a novel unsupervised approach to train projection networks for GANs, enhancing their utility in image processing tasks.

Findings

Effective image super-resolution using the projection network

Semantic clustering of images achieved without supervised labels

Applicable to pre-trained StyleGAN models

Abstract

We propose the use of unsupervised learning to train projection networks that project onto the latent space of an already trained generator. We apply our method to a trained StyleGAN, and use our projection network to perform image super-resolution and clustering of images into semantically identifiable groups.