Deep Automodulators

TL;DR

This paper introduces automodulators, a new type of generative autoencoder capable of high-quality image reconstruction and style-mixing, with novel training techniques demonstrated on multiple datasets.

Contribution

It presents the automodulator architecture, enabling faithful image reproduction and style-mixing, along with a training method for high-resolution, sharp outputs in autoencoders.

Findings

State-of-the-art autoencoder performance

High-resolution, sharp image outputs

Effective style-mixing capabilities

Abstract

We introduce a new category of generative autoencoders called automodulators. These networks can faithfully reproduce individual real-world input images like regular autoencoders, but also generate a fused sample from an arbitrary combination of several such images, allowing instantaneous 'style-mixing' and other new applications. An automodulator decouples the data flow of decoder operations from statistical properties thereof and uses the latent vector to modulate the former by the latter, with a principled approach for mutual disentanglement of decoder layers. Prior work has explored similar decoder architecture with GANs, but their focus has been on random sampling. A corresponding autoencoder could operate on real input images. For the first time, we show how to train such a general-purpose model with sharp outputs in high resolution, using novel training techniques, demonstrated on four image data sets. Besides style-mixing, we show state-of-the-art results in autoencoder comparison, and visual image quality nearly indistinguishable from state-of-the-art GANs. We expect the automodulator variants to become a useful building block for image applications and other data domains.