Open wusai80 opened 6 years ago
It’s output from our pretrained painting style estimator. The model was trained on wikiart dataset with 80,000 paintings. The model will be released in next update. You can also train your own too since the wikiart images are available online.
Because paintings have different styles and genres, using one set of parameters for all the paintings won't work well. Some are relatively photorealistic while the others are strongly stylized. The attached screenshot is one page of our supplemental document which gives details about what type of painting styles we used and the weights accordingly.
It seems to define many weight parameters. These weights are used to tune the style weights? How can we decide the value for a new image? Thanks.