Unsupervised Image-to-Image Translation with Stacked Cycle-Consistent Adversarial Networks
This addresses a bottleneck in image-to-image translation for computer vision applications, but it is incremental as it builds on existing cycle-consistent adversarial networks.
The paper tackles the problem of inferior results in unsupervised image-to-image translation at high resolutions or with significant domain differences by proposing Stacked Cycle-Consistent Adversarial Networks (SCANs), which decompose translation into multi-stage transformations and improve translation quality on multiple datasets.
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.