Delta-GAN-Encoder: Encoding Semantic Changes for Explicit Image Editing, using Few Synthetic Samples

TL;DR

This paper introduces Delta-GAN-Encoder, a novel autoencoder-based method that learns to encode semantic changes in GAN latent spaces for precise, controllable image editing using minimal synthetic samples, without requiring latent space structure.

Contribution

It presents a new approach for semantic change encoding in GANs that works with minimal data and no latent space assumptions, enabling high-quality image editing.

Findings

Achieves state-of-the-art results in facial image editing.

Requires only few synthetic samples for training.

Does not depend on latent space linearity or structure.

Abstract

Understating and controlling generative models' latent space is a complex task. In this paper, we propose a novel method for learning to control any desired attribute in a pre-trained GAN's latent space, for the purpose of editing synthesized and real-world data samples accordingly. We perform Sim2Real learning, relying on minimal samples to achieve an unlimited amount of continuous precise edits. We present an Autoencoder-based model that learns to encode the semantics of changes between images as a basis for editing new samples later on, achieving precise desired results - example shown in Fig. 1. While previous editing methods rely on a known structure of latent spaces (e.g., linearity of some semantics in StyleGAN), our method inherently does not require any structural constraints. We demonstrate our method in the domain of facial imagery: editing different expressions, poses, and lighting attributes, achieving state-of-the-art results.