Unsupervised Image-to-Image Translation with Stacked Cycle-Consistent Adversarial Networks

TL;DR

This paper introduces Stacked Cycle-Consistent Adversarial Networks (SCANs), a multi-stage approach that enhances unsupervised image-to-image translation quality, especially at high resolutions and with significant domain differences, by decomposing translation into coarse-to-fine steps.

Contribution

The paper proposes a novel multi-stage framework with adaptive fusion blocks to improve high-resolution unsupervised image translation over existing single-stage methods.

Findings

Improved translation quality on multiple datasets.

Effective high-resolution image translation in a coarse-to-fine manner.

Adaptive fusion blocks enhance information integration between stages.

Abstract

Recent studies on unsupervised image-to-image translation have made a remarkable progress by training a pair of generative adversarial networks with a cycle-consistent loss. However, such unsupervised methods may generate inferior results when the image resolution is high or the two image domains are of significant appearance differences, such as the translations between semantic layouts and natural images in the Cityscapes dataset. In this paper, we propose novel Stacked Cycle-Consistent Adversarial Networks (SCANs) by decomposing a single translation into multi-stage transformations, which not only boost the image translation quality but also enable higher resolution image-to-image translations in a coarse-to-fine manner. Moreover, to properly exploit the information from the previous stage, an adaptive fusion block is devised to learn a dynamic integration of the current stage's output and the previous stage's output. Experiments on multiple datasets demonstrate that our proposed approach can improve the translation quality compared with previous single-stage unsupervised methods.