Recreate photos in in the style of famous artists with machine learning. CSCI3341 Artificial Intelligence Final Project.
Make sure to download models with the util/download_models.sh
script first. Thanks @jcjohnson!
python art.py --help
Create a (low quality) version of the image below. The first positional argument is the content image and the image(s) after are the style images. The further options instruct the algorithm to run for 200 iterations and output an image with a 256px maximum width, which should be doable relatively quickly on a personal computer.
python art.py examples/gasson.jpg examples/starry.jpg -n 200 -w 256
Further options include using the entire oeuvre of an artist as a stylistic target (which never works well), scaling the stylistic features by a factor, specifying whether to initialize the candidate image from random noise or the content image, and more. See the script's help for details.
The best Gassongram ever:
util/wikiart-scraper.py
contains a neat script which will automatically scrape every painting for a given artist from wikiart. Simply specify the URL component of the artist's name (e.g. pablo-picasso
) in the ARTISTS
array in the script.