Semi Few-Shot Attribute Translation

TL;DR

This paper introduces a novel GAN-based meta-learning approach for few-shot image-to-image translation, enabling attribute transfer with limited labeled data, demonstrated on hair color synthesis tasks.

Contribution

It proposes a new meta-training method for GANs that facilitates effective attribute translation with very few labeled samples, advancing few-shot generative transfer learning.

Findings

Effective attribute transfer with few samples demonstrated on hair color tasks

Meta-trained GANs outperform traditional methods in low-data scenarios

Paves the way for further research in generative transfer learning

Abstract

Recent studies have shown remarkable success in image-to-image translation for attribute transfer applications. However, most of existing approaches are based on deep learning and require an abundant amount of labeled data to produce good results, therefore limiting their applicability. In the same vein, recent advances in meta-learning have led to successful implementations with limited available data, allowing so-called few-shot learning. In this paper, we address this limitation of supervised methods, by proposing a novel approach based on GANs. These are trained in a meta-training manner, which allows them to perform image-to-image translations using just a few labeled samples from a new target class. This work empirically demonstrates the potential of training a GAN for few shot image-to-image translation on hair color attribute synthesis tasks, opening the door to further research on generative transfer learning.