Inverting The Generator Of A Generative Adversarial Network (II)

TL;DR

This paper introduces a method to invert pre-trained GANs, mapping images back to their latent space, enabling analysis of GAN capabilities and performance comparison across models.

Contribution

The paper presents a novel inversion technique for GANs that allows for latent space projection of images, facilitating performance evaluation and attribute analysis.

Findings

Effective latent space reconstruction of images.

Quantitative comparison of GAN models.

Insights into dataset attributes modeled by GANs.

Abstract

Generative adversarial networks (GANs) learn a deep generative model that is able to synthesise novel, high-dimensional data samples. New data samples are synthesised by passing latent samples, drawn from a chosen prior distribution, through the generative model. Once trained, the latent space exhibits interesting properties, that may be useful for down stream tasks such as classification or retrieval. Unfortunately, GANs do not offer an "inverse model", a mapping from data space back to latent space, making it difficult to infer a latent representation for a given data sample. In this paper, we introduce a technique, inversion, to project data samples, specifically images, to the latent space using a pre-trained GAN. Using our proposed inversion technique, we are able to identify which attributes of a dataset a trained GAN is able to model and quantify GAN performance, based on a reconstruction loss. We demonstrate how our proposed inversion technique may be used to quantitatively compare performance of various GAN models trained on three image datasets. We provide code for all of our experiments, https://github.com/ToniCreswell/InvertingGAN.