Technical Aspects

Many of the video clips shown here are based on an algorithm, which can transfer a picture’s painting style to another image using a convolutional neural network. The algorithm was first presented in 2015 in a paper by Leon A. Gatys, Alexander S. Ecker and Matthias Bethge. The authors discuss techniques to train a neural network in such a way that the artistic style of a painting can be separated from the image structure and then combined with any other image.

Justin Johnson provided a torch implementation of the algorithm on GitHub. Processing the video clips shown here required a workstation with an i7-6700K CPU and a GeForce GTX Titan X graphics card with 12 GB RAM running on Ubuntu 15.04. Using a CUDA-capable graphics card greatly speeds up the processing time.

Installation of neural-style
Working with neural-style
Style Transfer for Video Clips
Post-production
Alternative Techniques