One-Shot Unsupervised Cross Domain Translation

TL;DR

This paper introduces a novel one-shot unsupervised cross-domain translation method that adapts a variational autoencoder to generate domain B equivalents of a single image from domain A, outperforming existing methods trained on multiple samples.

Contribution

The paper proposes a new two-step variational autoencoder approach for one-shot domain translation, enabling effective cross-domain mapping with only a single sample from domain A.

Findings

Performs comparably to multi-sample methods on one-shot tasks

Fails to outperform existing methods on multi-sample training

Code is publicly available for reproducibility

Abstract

Given a single image x from domain A and a set of images from domain B, our task is to generate the analogous of x in B. We argue that this task could be a key AI capability that underlines the ability of cognitive agents to act in the world and present empirical evidence that the existing unsupervised domain translation methods fail on this task. Our method follows a two step process. First, a variational autoencoder for domain B is trained. Then, given the new sample x, we create a variational autoencoder for domain A by adapting the layers that are close to the image in order to directly fit x, and only indirectly adapt the other layers. Our experiments indicate that the new method does as well, when trained on one sample x, as the existing domain transfer methods, when these enjoy a multitude of training samples from domain A. Our code is made publicly available at https://github.com/sagiebenaim/OneShotTranslation