UNIT
UNIT copied to clipboard
Unsupervised Image-to-Image Translation
UNIT: UNsupervised Image-to-image Translation Networks
New implementation available at imaginaire repository
We have a reimplementation of the UNIT method that is more performant. It is avaiable at Imaginaire
License
Copyright (C) 2018 NVIDIA Corporation. All rights reserved. Licensed under the CC BY-NC-SA 4.0 license (https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode).
Code usage
-
Please check out our tutorial.
-
For multimodal (or many-to-many) image translation, please check out our new work on MUNIT.
What's new.
-
05-02-2018: We now adapt MUNIT code structure. For reproducing experiment results in the NIPS paper, please check out version_02 branch.
-
12-21-2017: Release pre-trained synthia-to-cityscape image translation model. See USAGE.md for usage examples.
-
12-14-2017: Added multi-scale discriminators described in the pix2pixHD paper. To use it simply make the name of the discriminator COCOMsDis.
Paper
Two Minute Paper Summary
(We thank the Two Minute Papers channel for summarizing our work.)
The Shared Latent Space Assumption
Result Videos
More image results are available in the Google Photo Album.
Left: input. Right: neural network generated. Resolution: 640x480
Left: input. Right: neural network generated. Resolution: 640x480
- Snowy2Summery-01
- Snowy2Summery-02
- Day2Night-01
- Day2Night-02
- Translation Between 5 dog breeds
- Translation Between 6 cat species
Street Scene Image Translation
From the first row to the fourth row, we show example results on day to night, sunny to rainy, summery to snowy, and real to synthetic image translation (two directions).
For each image pair, left is the input image; right is the machine generated image.