Geometric Portrait Stylization

With this implementation you can generate geometric stylizations of portraits and scenic input images as described in the paper.

Dependencies

Used libraries can be easily installed with pip.

Required

• NumPy for vectorized arrays.
• OpenCV for image processing.
• dlib for face landmark detection.

The dlib face detector also requires a pretrained model for the landmark detection. If you have the requests lib installed, it will download the model automatically. Otherwise, you can lookup the url and target path in the util.py file and download the model manually.

Further common utility methods are bundled in my drbutil which you can also install as a library using pip install drbutil. However, if not installed, it will be downloaded automatically and imported from source.

Run Examples

• In the main directory you can run python runExamples.py to generate example results. This script contains exemplary setups to recreate results from the paper and an explanation of the used parameters. Here you can also set the result directory and if you want to show previews during the computation.
• Results are stored as .png and .svg files, respectively.

Citation

You can cite the paper with:

@inproceedings{bukenberger2024geometric,
	booktitle = {Vision, Modeling, and Visualization},
	editor = {Linsen, Lars and Thies, Justus},
	title = {{Geometric Portrait Stylization}},
	author = {Bukenberger, Dennis R.},
	year = {2024},
	publisher = {The Eurographics Association},
	ISBN = {978-3-03868-247-9},
	DOI = {10.2312/vmv.20241203}
}