Optimizing Latent Space Directions For GAN-based Local Image Editing

TL;DR

This paper introduces LELSD, a new method for localized image editing with GANs that improves edit locality, disentanglement, and computational efficiency by leveraging a pre-trained segmentation network.

Contribution

The paper proposes a novel objective function and a framework that enhances local image editing in GANs, applicable across datasets and architectures, with improved disentanglement and speed.

Findings

High-quality localized edits on GAN-generated images

Effective editing on real images

Fast computation and high disentanglement

Abstract

Generative Adversarial Network (GAN) based localized image editing can suffer from ambiguity between semantic attributes. We thus present a novel objective function to evaluate the locality of an image edit. By introducing the supervision from a pre-trained segmentation network and optimizing the objective function, our framework, called Locally Effective Latent Space Direction (LELSD), is applicable to any dataset and GAN architecture. Our method is also computationally fast and exhibits a high extent of disentanglement, which allows users to interactively perform a sequence of edits on an image. Our experiments on both GAN-generated and real images qualitatively demonstrate the high quality and advantages of our method.