Interpreting Generative Adversarial Networks for Interactive Image Generation

TL;DR

This paper reviews recent methods for interpreting GANs, focusing on understanding how they generate realistic images from random vectors and enabling interactive image editing.

Contribution

It categorizes interpretation techniques into supervised, unsupervised, and embedding-guided approaches, highlighting how human-understandable concepts emerge in GAN representations.

Findings

Interpretation methods help understand GAN image generation.

Human-understandable concepts can be identified in GAN representations.

These concepts enable interactive image editing.

Abstract

Significant progress has been made by the advances in Generative Adversarial Networks (GANs) for image generation. However, there lacks enough understanding of how a realistic image is generated by the deep representations of GANs from a random vector. This chapter gives a summary of recent works on interpreting deep generative models. The methods are categorized into the supervised, the unsupervised, and the embedding-guided approaches. We will see how the human-understandable concepts that emerge in the learned representation can be identified and used for interactive image generation and editing.