A Semi-Paired Approach For Label-to-Image Translation

TL;DR

This paper introduces a semi-supervised framework for label-to-image translation that effectively utilizes limited paired data and larger unpaired datasets, improving photorealistic image generation from semantic labels.

Contribution

It presents the first semi-paired approach for label-to-image translation, leveraging input reconstruction and a rare classes sampling algorithm to enhance performance with minimal paired data.

Findings

Outperforms state-of-the-art unsupervised and semi-supervised methods

Achieves comparable or better results than fully supervised models with fewer paired samples

Demonstrates effectiveness across three standard benchmarks

Abstract

Data efficiency, or the ability to generalize from a few labeled data, remains a major challenge in deep learning. Semi-supervised learning has thrived in traditional recognition tasks alleviating the need for large amounts of labeled data, yet it remains understudied in image-to-image translation (I2I) tasks. In this work, we introduce the first semi-supervised (semi-paired) framework for label-to-image translation, a challenging subtask of I2I which generates photorealistic images from semantic label maps. In the semi-paired setting, the model has access to a small set of paired data and a larger set of unpaired images and labels. Instead of using geometrical transformations as a pretext task like previous works, we leverage an input reconstruction task by exploiting the conditional discriminator on the paired data as a reverse generator. We propose a training algorithm for this shared network, and we present a rare classes sampling algorithm to focus on under-represented classes. Experiments on 3 standard benchmarks show that the proposed model outperforms state-of-the-art unsupervised and semi-supervised approaches, as well as some fully supervised approaches while using a much smaller number of paired samples.