Conditional Invertible Neural Networks for Diverse Image-to-Image Translation

TL;DR

This paper presents a conditional invertible neural network (cINN) architecture for diverse and controllable image-to-image translation, leveraging invertible models' advantages like mode collapse immunity, demonstrated on tasks like day-night translation and colorization.

Contribution

Introduces a novel cINN architecture combining INNs with feed-forward networks for efficient, diverse image translation with stable training and latent space manipulation capabilities.

Findings

cINNs enable diverse image translation with stable training.

The model demonstrates effective day-night translation and colorization.

Latent space manipulation allows intuitive style changes.

Abstract

We introduce a new architecture called a conditional invertible neural network (cINN), and use it to address the task of diverse image-to-image translation for natural images. This is not easily possible with existing INN models due to some fundamental limitations. The cINN combines the purely generative INN model with an unconstrained feed-forward network, which efficiently preprocesses the conditioning image into maximally informative features. All parameters of a cINN are jointly optimized with a stable, maximum likelihood-based training procedure. Even though INN-based models have received far less attention in the literature than GANs, they have been shown to have some remarkable properties absent in GANs, e.g. apparent immunity to mode collapse. We find that our cINNs leverage these properties for image-to-image translation, demonstrated on day to night translation and image colorization. Furthermore, we take advantage of our bidirectional cINN architecture to explore and manipulate emergent properties of the latent space, such as changing the image style in an intuitive way.