From Continuity to Editability: Inverting GANs with Consecutive Images

TL;DR

This paper introduces a novel GAN inversion method using consecutive images, which enhances both image reconstruction fidelity and editability, and supports video-based inversion and semantic transfer.

Contribution

It leverages the continuity of consecutive images to regularize inversion, improving both fidelity and editability, and is the first to support video-based GAN inversion.

Findings

Outperforms state-of-the-art in fidelity and editability

Supports video-based GAN inversion

Enables unsupervised semantic transfer

Abstract

Existing GAN inversion methods are stuck in a paradox that the inverted codes can either achieve high-fidelity reconstruction, or retain the editing capability. Having only one of them clearly cannot realize real image editing. In this paper, we resolve this paradox by introducing consecutive images (\eg, video frames or the same person with different poses) into the inversion process. The rationale behind our solution is that the continuity of consecutive images leads to inherent editable directions. This inborn property is used for two unique purposes: 1) regularizing the joint inversion process, such that each of the inverted code is semantically accessible from one of the other and fastened in a editable domain; 2) enforcing inter-image coherence, such that the fidelity of each inverted code can be maximized with the complement of other images. Extensive experiments demonstrate that our alternative significantly outperforms state-of-the-art methods in terms of reconstruction fidelity and editability on both the real image dataset and synthesis dataset. Furthermore, our method provides the first support of video-based GAN inversion, and an interesting application of unsupervised semantic transfer from consecutive images. Source code can be found at: \url{https://github.com/cnnlstm/InvertingGANs_with_ConsecutiveImgs}.