Smoothing the Disentangled Latent Style Space for Unsupervised Image-to-image Translation
Yahui Liu1,3, Enver Sangineto1, Yajing Chen2, Linchao Bao2, Haoxian Zhang2, Nicu Sebe1, Bruno Lepri3, Wei Wang1, Marco De Nadai3
1University of Trento, Italy, 2Tencent AI Lab, China, 3Bruno Kessler Foundation, Italy.
To appear in CVPR 2021
The repository offers the official implementation of our paper in PyTorch.

Image-to-Image (I2I) multi-domain translation models are usually evaluated also using the quality of their semantic interpolation results. However, state-of-the-art models frequently show abrupt changes in the image appearance during interpolation, and usually perform poorly in interpolations across domains. In this paper, we propose a new training protocol based on three specific losses which help a translation network to learn a smooth and disentangled latent style space in which: 1) Both intra- and inter-domain interpolations correspond to gradual changes in the generated images and 2) The content of the source image is better preserved during the translation. Moreover, we propose a novel evaluation metric to properly measure the smoothness of latent style space of I2I translation models. The proposed method can be plugged in existing translation approaches, and our extensive experiments on different datasets show that it can significantly boost the quality of the generated images and the graduality of the interpolations.


Our method generates smooth interpolations within and across domains in various image-to-image translation tasks.

Teaser video

Click the figure to watch the teaser video.



See the environment.yml. We provide an user-friendly configuring method via Conda system, and you can create a new Conda environment using the command:

conda env create -f environment.yml

Codes will be released soon ...


For fast testing, we provide pretrained models on CelebA-HQ (gender) and AFHQ (animal faces):

GoogleDrive GoogleDrive

The models can be tested directly by using the offical codes of StarGAN v2.


  • Data Preparing

  • Training


This code is based on the StarGAN v2. Thanks to the contributors of this project.


    author    = {Liu, Yahui and Sangineto, Enver and Chen, Yajing and Bao, Linchao and Zhang, Haoxian and Sebe, Nicu and Lepri, Bruno and Wang, Wei and De Nadai, Marco},
    title     = {Smoothing the Disentangled Latent Style Space for Unsupervised Image-to-Image Translation},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2021},
    pages     = {10785-10794}

If you have any questions, please contact me without hesitation (yahui.liu AT