When I was looking for examples of style transfer in TensorFlow I found a lot of repos that were overcomplicated with many files and working parts. Rather than having extensible and "use" ready code I wanted code that was just concise and easy to read so I could understand style transfer.
So rather than have a million configurable command line arguments and a bunch
of utility files there's just one main.py
taking the simpliest approach to
style transfer with comments and explanations all in under 150 lines.
vgg19.py
simply defines a pretrained VGG network. You need to download the
VGG weights at http://www.vlfeat.org/matconvnet/models/imagenet-vgg-verydeep-19.mat.
See the accompanying blog post for this at https://www.andrewszot.com/blog/machine_learning/deep_learning/style_transfer.