NAM: Non-Adversarial Unsupervised Domain Mapping

TL;DR

NAM introduces a non-adversarial approach for unsupervised image domain translation that leverages pre-trained generative models, resulting in more stable training and higher quality translations compared to adversarial methods.

Contribution

The paper proposes NAM, a novel non-adversarial method that separates generative modeling from domain mapping, improving stability and translation quality.

Findings

Achieves higher resolution image translations.

Offers simpler and more stable training process.

Demonstrates superior results in experiments.

Abstract

Several methods were recently proposed for the task of translating images between domains without prior knowledge in the form of correspondences. The existing methods apply adversarial learning to ensure that the distribution of the mapped source domain is indistinguishable from the target domain, which suffers from known stability issues. In addition, most methods rely heavily on `cycle' relationships between the domains, which enforce a one-to-one mapping. In this work, we introduce an alternative method: Non-Adversarial Mapping (NAM), which separates the task of target domain generative modeling from the cross-domain mapping task. NAM relies on a pre-trained generative model of the target domain, and aligns each source image with an image synthesized from the target domain, while jointly optimizing the domain mapping function. It has several key advantages: higher quality and resolution image translations, simpler and more stable training and reusable target models. Extensive experiments are presented validating the advantages of our method.