Finding an Unsupervised Image Segmenter in Each of Your Deep Generative Models

TL;DR

This paper introduces an unsupervised method to find image segmentation directions in GAN latent spaces, enabling effective foreground-background separation without human supervision across various models and datasets.

Contribution

It presents a generator-agnostic, automatic procedure for discovering segmentation directions in GANs, eliminating the need for human-labeled data or training on specific datasets.

Findings

Achieves strong segmentation results across multiple GAN architectures.

Performs well on standard image segmentation benchmarks.

Does not require access to training data or human supervision.

Abstract

Recent research has shown that numerous human-interpretable directions exist in the latent space of GANs. In this paper, we develop an automatic procedure for finding directions that lead to foreground-background image separation, and we use these directions to train an image segmentation model without human supervision. Our method is generator-agnostic, producing strong segmentation results with a wide range of different GAN architectures. Furthermore, by leveraging GANs pretrained on large datasets such as ImageNet, we are able to segment images from a range of domains without further training or finetuning. Evaluating our method on image segmentation benchmarks, we compare favorably to prior work while using neither human supervision nor access to the training data. Broadly, our results demonstrate that automatically extracting foreground-background structure from pretrained deep generative models can serve as a remarkably effective substitute for human supervision.