ST-VAE

Multiple style transfer via variational autoencoder

By Zhi-Song Liu, Vicky Kalogeiton and Marie-Paule Cani

This repo only provides simple testing codes, pretrained models and the network strategy demo.

We propose a Multiple style transfer via variational autoencoder (ST-VAE)

Please check our paper or arxiv paper

BibTex

    @InProceedings{Liu2021stvae,
        author = {Zhi-Song Liu and Wan-Chi Siu and Marie-Paule Cani},
        title = {Multiple Style Transfer via Variational AutoEncoder},
        booktitle = {2021 IEEE International Conference on Image Processing(ICIP)},
        month = {Oct},
        year = {2021}
    }

For proposed ST-VAE model, we claim the following points:

• First working on using Variational AutoEncoder for image style transfer.

• Multiple style transfer by proposed VAE based Linear Transformation.

Dependencies

Python > 3.0
Pytorch > 1.0
NVIDIA GPU + CUDA

Complete Architecture

The complete architecture is shown as follows,

Visualization

1. Single style transfer

2. Multiple style transfer

Implementation

1. Quick testing

1. Download pre-trained models from

https://drive.google.com/file/d/1WZrvjCGBO1mpggkdJiaw8jp-6ywbXn4J/view?usp=sharing

and copy them to the folder "models"

2. Put your content image under "Test/content" and your style image under "Test/style"

3. For single style transfer, run

$ python eval.py 

The stylized images will be in folder "Test/result" 4. For multiple style transfer, run

$ python eval_multiple_style.py

5. For real-time demo, run

$ python real-time-demo.py --style_image Test/style/picasso_self_portrait.jpg

6. For training, put the training images under the folder "train_data"

download MS-COCO dataset from https://cocodataset.org/#home and put it under "train_data/content" download Wikiart from https://www.wikiart.org/ and put them under "train_data/style" then run,

$ python train.py

Special thanks to the contributions of Jakub M. Tomczak for their LT on their LT computation