This is an implementation of 'A Neural Algorithm of Artistic Style' (http://arxiv.org/abs/1508.06576). The code is largely based on https://github.com/pfnet-research/chainer-gogh and https://github.com/yusuketomoto/chainer-fast-neuralstyle
The script generates a stylised image from an input image and a style image.
Demo on Google Colaboratory
You can try the demo on a browser Google Colaboratory
- Python 3: Anaconda is recommended
- Python libraries: Chainer, cupy:
pip install cupy chainer
- CUDA supported GPU is highly recommended. Without one, it takes ages to do anything with CNN.
How to use
python train.py -h
gives a brief description of command line arguments.
A typical conversion is performed by
python train_gogh.py input.jpg -s style.jpg -rs
Generated images will be found under the directory named 'result'.