Latent Filter Scaling for Multimodal Unsupervised Image-To-Image Translation

Abstract
In multimodal unsupervised image-to-image translation tasks, the goal is to translate an image from the source domain to many images in the target domain. We present a simple method that produces higher quality images than current state-of-the-art while maintaining the same amount of multimodal diversity. Previous methods follow the unconditional approach of trying to map the latent code directly to a full-size image. This leads to complicated network architectures with several introduced hyperparameters to tune. By treating the latent code as a modifier of the convolutional filters, we produce multimodal output while maintaining the traditional Generative Adversarial Network (GAN) loss and without additional hyperparameters. The only tuning required by our method controls the tradeoff between variability and quality of generated images. Furthermore, we achieve disentanglement between source domain content and target domain style for free as a by-product of our formulation. We perform qualitative and quantitative experiments showing the advantages of our method compared with the state-of-the art on multiple benchmark image-to-image translation datasets.

Citation
Alharbi, Y., Smith, N., & Wonka, P. (2019). Latent Filter Scaling for Multimodal Unsupervised Image-To-Image Translation. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). doi:10.1109/cvpr.2019.00155

Acknowledgements
The project was funded in part by the KAUST Office of Sponsored Research (OSR) under Award No. URF/1/3426-01-01.

Publisher
Institute of Electrical and Electronics Engineers (IEEE)

Conference/Event Name
2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

DOI
10.1109/CVPR.2019.00155

arXiv
1812.09877

Additional Links
https://ieeexplore.ieee.org/document/8953741/https://ieeexplore.ieee.org/document/8953741/https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8953741

Permanent link to this record

Version History

Now showing 1 - 2 of 2
VersionDateSummary
2*
2019-11-28 07:07:16
Conference paper released
2019-11-28 07:05:11
* Selected version