GANSpace: Discovering Interpretable GAN Controls

TL;DR

This paper introduces a simple PCA-based method to analyze GANs and create interpretable controls for image editing, enabling intuitive manipulation of generated images across various styles and datasets.

Contribution

It presents a novel PCA-based technique for discovering interpretable latent directions and demonstrates layer-wise control in BigGAN and StyleGAN models.

Findings

Effective layer-wise perturbation controls for image synthesis.

Good qualitative matches with supervised edit directions.

Works across different GAN architectures and datasets.

Abstract

This paper describes a simple technique to analyze Generative Adversarial Networks (GANs) and create interpretable controls for image synthesis, such as change of viewpoint, aging, lighting, and time of day. We identify important latent directions based on Principal Components Analysis (PCA) applied either in latent space or feature space. Then, we show that a large number of interpretable controls can be defined by layer-wise perturbation along the principal directions. Moreover, we show that BigGAN can be controlled with layer-wise inputs in a StyleGAN-like manner. We show results on different GANs trained on various datasets, and demonstrate good qualitative matches to edit directions found through earlier supervised approaches.