1.0.0

Transfer is the Style Transfer subnetwork. It takes as input a content tensor of the shape 960x1920x?? and a style_params vector of varying size and outputs the stylized image as a tensor of the shape 960x1920x3.

Predictor is the Style Prediction subnetwork. It takes as input a style tensor of the shape 960x1920x3 and outputs the style_params needed for the Style Transfer subnetwork.

Training are the TensorFlow models that were created via model.save(). They can be loaded into TensorFlow via Model.load(). They include the loss calculation as metrics, the style predictor and the style transfer subnetworks. They take as input a content tensor of the shape 960x1920x?? and a style tensor of the shape 960x1920x3 and outputs the stylized image as a tensor of the shape 960x1920x3.

For more details please read the paper and visit the website.