Multi-style Generative Network for Real-time Transfer [arXiv] [project] Hang Zhang, Kristin Dana @article{zhang2017multistyle, title={Multi-style Generative Network for Real-time Transfer}, author={Zhang, Hang and Dana, Kristin}, journal={arXiv preprint arXiv:1703.06953}, year={2017} }

Installation

We also provide PyTorch implementation and MXNet implementation. Please install Torch7 with cuda and cudnn support. The code has been tested on Ubuntu 16.04 with Titan X Pascal and Maxwell.

luarocks install https://raw.githubusercontent.com/zhanghang1989/MSG-Net/master/texture-scm-1.rockspec

Test and Demo

0. Clone the repo and download pre-trained models

   git clone git@github.com:zhanghang1989/MSG-Net.git
   cd MSG-Net/experiments
   bash models/download_models.sh 

1. Web Camera Demo

   qlua webcam.lua
   

   
2. Test on Image

   th test.lua -input_image images/content/venice-boat.jpg -image_size 1024
   eog stylized
   

[More Example Results]

Train Your Own Model

Please follow this tutorial to train a new model.

Release Timeline

• 03/20/2017 we have released the demo video.
• 03/24/2017 We have released ArXiv paper and test code with pre-trained models.
• 04/09/2017 We have released the training code.
• 04/24/2017 Please checkout our PyTorch implementation.

Acknowledgement

The code benefits from outstanding prior work and their implementations including:

• Texture Networks: Feed-forward Synthesis of Textures and Stylized Images by Ulyanov et al. ICML 2016. (code)
• Perceptual Losses for Real-Time Style Transfer and Super-Resolution by Johnson et al. ECCV 2016 (code)
• Image Style Transfer Using Convolutional Neural Networks by Gatys et al. CVPR 2016 and its torch implementation code by Johnson.