Latent Transformations via NeuralODEs for GAN-based Image Editing

TL;DR

This paper introduces a method using NeuralODEs to perform nonlinear latent space transformations, improving image editing capabilities beyond linear approaches especially for complex, non-face images.

Contribution

It proposes a novel approach employing NeuralODEs for nonlinear latent space manipulations, enhancing semantic image editing for diverse and complex datasets.

Findings

NeuralODE-based transformations outperform linear methods on complex datasets.

Certain attribute manipulations are difficult with linear shifts but feasible with nonlinear flows.

The method improves controllability in non-face image editing tasks.

Abstract

Recent advances in high-fidelity semantic image editing heavily rely on the presumably disentangled latent spaces of the state-of-the-art generative models, such as StyleGAN. Specifically, recent works show that it is possible to achieve decent controllability of attributes in face images via linear shifts along with latent directions. Several recent methods address the discovery of such directions, implicitly assuming that the state-of-the-art GANs learn the latent spaces with inherently linearly separable attribute distributions and semantic vector arithmetic properties. In our work, we show that nonlinear latent code manipulations realized as flows of a trainable Neural ODE are beneficial for many practical non-face image domains with more complex non-textured factors of variation. In particular, we investigate a large number of datasets with known attributes and demonstrate that certain attribute manipulations are challenging to obtain with linear shifts only.