Learning to Transfer: Unsupervised Meta Domain Translation

TL;DR

This paper introduces MT-GAN, a meta-learning approach for unsupervised domain translation that learns a good model initialization, enabling effective adaptation to new domains with minimal data.

Contribution

The paper proposes a novel meta-learning framework for unsupervised domain translation, improving adaptability and performance on small datasets compared to existing methods.

Findings

Outperforms existing methods on ten diverse translation tasks

Effective with as few as ten training samples per domain

Demonstrates strong generalization ability across multiple face translation tasks

Abstract

Unsupervised domain translation has recently achieved impressive performance with Generative Adversarial Network (GAN) and sufficient (unpaired) training data. However, existing domain translation frameworks form in a disposable way where the learning experiences are ignored and the obtained model cannot be adapted to a new coming domain. In this work, we take on unsupervised domain translation problems from a meta-learning perspective. We propose a model called Meta-Translation GAN (MT-GAN) to find good initialization of translation models. In the meta-training procedure, MT-GAN is explicitly trained with a primary translation task and a synthesized dual translation task. A cycle-consistency meta-optimization objective is designed to ensure the generalization ability. We demonstrate effectiveness of our model on ten diverse two-domain translation tasks and multiple face identity translation tasks. We show that our proposed approach significantly outperforms the existing domain translation methods when each domain contains no more than ten training samples.