During the first month of my PhD I worked on the style transfer technique applied to videos. The objective is transform the news into a “cartoonized” video for kids by transferring the style of a cartoon.

I encapsulated the code of cysmith in a GPU-ready image with all the dependencies, including the download of the VGG-19 pretrained weights.

The style transfer process requires 2 photos:

The following video transfer the style from the “The Powerpuff Girls” to the weather forecast:


Having Docker, Docker Compose, NVIDIA-Docker and a NVIDIA GPU run:

docker-compose -f docker/docker-compose.yml run neuralstyle bash

Then, use the neural_style.py script on images/videos:

python neural_style.py --content_img golden_gate.jpg \
                       --style_imgs starry-night.jpg \
                       --max_size 500 \
                       --max_iterations 100 \


This repository is a Dockerized version of https://github.com/cysmith/neural-style-tf