Awesome

Universal Style Transfer

This is the Pytorch implementation of Universal Style Transfer via Feature Transforms.

Official Torch implementation can be found here and Tensorflow implementation can be found here.

Prerequisites

• Pytorch
• torchvision
• Pretrained encoder and decoder models for image reconstruction only (download and uncompress them under models/)
• CUDA + CuDNN

Prepare images

Simply put content and image pairs in images/content and images/style respectively. Note that correspoding conternt and image pairs should have same names.

Style Transfer

python WCT.py --cuda

Results

<img src="images/content/in1.jpg" width="200" hspace="5"><img src="images/style/in1.jpg" width="200" hspace="5"><img src="images/content/in3.jpg" width="200" hspace="5"><img src="images/style/in3.jpg" width="200" hspace="5">

<img src="samples/in1.jpg" width="400" hspace="10"><img src="samples/in3.jpg" width="400" hspace="10">

<img src="images/content/in2.jpg" width="200" hspace="5"><img src="images/style/in2.jpg" width="200" hspace="5"><img src="images/content/in4.jpg" width="200" hspace="5"><img src="images/style/in4.jpg" width="200" hspace="5">

<img src="samples/in2.jpg" width="400" hspace="10"><img src="samples/in4.jpg" width="400" hspace="10">

Acknowledgments

Many thanks to the author Yijun Li for his kind help.

Reference

Li Y, Fang C, Yang J, et al. Universal Style Transfer via Feature Transforms[J]. arXiv preprint arXiv:1705.08086, 2017.