Rethinking Style Transfer: From Pixels to Parameterized Brushstrokes.
Dmytro Kotovenko*, Matthias Wright*, Arthur Heimbrecht, and Björn Ommer.
* denotes equal contribution

We provide implementations in Tensorflow 1 and Tensorflow 2. In order to reproduce the results from the paper, we recommend the Tensorflow 1 implementation.

1. Clone this repository:

   > git clone https://github.com/CompVis/brushstroke-parameterized-style-transfer
   > cd brushstroke-parameterized-style-transfer

2. Install Tensorflow 1.14 (preferably with GPU support).
   If you are using Conda, this command will create a new environment and install Tensorflow as well as compatible CUDA and cuDNN versions.

   > conda create --name tf14 tensorflow-gpu==1.14
   > conda activate tf14

3. Install requirements:

   > pip install -r requirements.txt

from PIL import Image
import model

content_img = Image.open('images/content/golden_gate.jpg')
style_img = Image.open('images/style/van_gogh_starry_night.jpg')

stylized_img = model.stylize(content_img,
                             style_img,
                             num_strokes=5000,
                             num_steps=100,
                             content_weight=1.0,
                             style_weight=3.0,
                             num_steps_pixel=1000)

stylized_img.save('images/stylized.jpg')

or open Colab.

We created a Streamlit app where you can draw curves to control the flow of brushstrokes.

To run the app on your own machine:

> CUDA_VISIBLE_DEVICES=0 streamlit run app.py

You can also run the app on a remote server and forward the port to your local machine: https://docs.streamlit.io/en/0.66.0/tutorial/run_streamlit_remotely.html

If you don't have access to GPUs we also created a Colab from which you can start the drawing app.

PyTorch implementation by justanhduc.

@article{kotovenko_cvpr_2021,
    title={Rethinking Style Transfer: From Pixels to Parameterized Brushstrokes},
    author={Dmytro Kotovenko and Matthias Wright and Arthur Heimbrecht and Bj{\"o}rn Ommer},
    journal={CVPR},
    year={2021}
}