GAN Inversion for Image Editing via Unsupervised Domain Adaptation

TL;DR

This paper introduces UDA-inversion, a novel unsupervised domain adaptation method for GAN inversion that effectively reconstructs and edits both high-quality and low-quality images, improving practical applicability.

Contribution

It proposes a theoretical framework with an upper bound on loss, enabling unsupervised learning of representations for both HQ and LQ images in GAN inversion.

Findings

Achieves a PSNR of 22.14 on FFHQ dataset.

Performs comparably to supervised methods.

Provides a theoretical guarantee for domain adaptation in GAN inversion.

Abstract

Existing GAN inversion methods work brilliantly in reconstructing high-quality (HQ) images while struggling with more common low-quality (LQ) inputs in practical application. To address this issue, we propose Unsupervised Domain Adaptation (UDA) in the inversion process, namely UDA-inversion, for effective inversion and editing of both HQ and LQ images. Regarding unpaired HQ images as the source domain and LQ images as the unlabeled target domain, we introduce a theoretical guarantee: loss value in the target domain is upper-bounded by loss in the source domain and a novel discrepancy function measuring the difference between two domains. Following that, we can only minimize this upper bound to obtain accurate latent codes for HQ and LQ images. Thus, constructive representations of HQ images can be spontaneously learned and transformed into LQ images without supervision. UDA-Inversion achieves a better PSNR of 22.14 on FFHQ dataset and performs comparably to supervised methods.