One-Sided Unsupervised Domain Mapping

TL;DR

This paper introduces a novel one-sided unsupervised domain mapping method that learns mappings without requiring inverse mappings, by preserving distances between samples and parts of samples, leading to improved results.

Contribution

The work proposes a distance-preserving approach for one-sided unsupervised domain mapping, eliminating the need for learning inverse mappings and outperforming circularity-based methods.

Findings

Effective one-sided mapping learned without inverse models

Maintains distances between samples and within samples

Achieves better numerical results than existing methods

Abstract

In unsupervised domain mapping, the learner is given two unmatched datasets $A$ and $B$. The goal is to learn a mapping $G_{AB}$ that translates a sample in $A$ to the analog sample in $B$. Recent approaches have shown that when learning simultaneously both $G_{AB}$ and the inverse mapping $G_{BA}$, convincing mappings are obtained. In this work, we present a method of learning $G_{AB}$ without learning $G_{BA}$. This is done by learning a mapping that maintains the distance between a pair of samples. Moreover, good mappings are obtained, even by maintaining the distance between different parts of the same sample before and after mapping. We present experimental results that the new method not only allows for one sided mapping learning, but also leads to preferable numerical results over the existing circularity-based constraint. Our entire code is made publicly available at https://github.com/sagiebenaim/DistanceGAN .