Linear Semantics in Generative Adversarial Networks

TL;DR

This paper reveals that GANs encode image semantics linearly in their internal features, enabling simple semantic control and few-shot image editing with minimal labeled data.

Contribution

It demonstrates that a linear transformation of GAN feature maps can extract semantics, facilitating semantic control and few-shot editing methods.

Findings

Semantic information is linearly encoded in GAN features.

Few-shot semantic segmentation model can be learned from 8 labeled images.

Proposed methods enable semantic control with minimal annotations.

Abstract

Generative Adversarial Networks (GANs) are able to generate high-quality images, but it remains difficult to explicitly specify the semantics of synthesized images. In this work, we aim to better understand the semantic representation of GANs, and thereby enable semantic control in GAN's generation process. Interestingly, we find that a well-trained GAN encodes image semantics in its internal feature maps in a surprisingly simple way: a linear transformation of feature maps suffices to extract the generated image semantics. To verify this simplicity, we conduct extensive experiments on various GANs and datasets; and thanks to this simplicity, we are able to learn a semantic segmentation model for a trained GAN from a small number (e.g., 8) of labeled images. Last but not least, leveraging our findings, we propose two few-shot image editing approaches, namely Semantic-Conditional Sampling and Semantic Image Editing. Given a trained GAN and as few as eight semantic annotations, the user is able to generate diverse images subject to a user-provided semantic layout, and control the synthesized image semantics. We have made the code publicly available.