neginamini / lotr-face-style-transfer

LOTR-Face-style-transfer

The demo for this project is a colab notebook, showing some results and example usage. In order to run the algorithm on a local machine, please run the jupyter notebook or, from command line, run the python script main.py (after installing requirements contained in requirements.txt), with the line:

python main.py --content_image path_to_content --style_image path_to_style

Other arguments to be used for obtaining better results are --content_weight, --style_weight and --regularization_weight, for which example values can be found inside the notebook, and in general can and should be tuned with respect to intensity of the selected pair of content/style images.

Reference works

This work is largely based on the Automated deep photo style transfer tensorflow implementation, for doing the conditional style tranfer on parsed faces. The face parsing was done using the BiSeNet network, in the implementation of zllrunning.