This is a re-working of the code found here, set up to use Docker and Jupyter notebooks.

This repository uses TensorFlow to model artistic style transfer. TensorFlow is fairly simple to install, but if you don't wish to install it directly on your system, you may run it using containers.

To get started for the first time, install Docker. Then, create a docker-machine. If you haven't done this before, and you are working on OSX, you can try to use the start-docker-osx.sh script:

$ source start-docker-osx.sh

Note that this script was developed for use with Docker Toolbox and may not work with the latest versions of Docker for OSX.

Once the Docker machine exists (here, named 'neuralart'), we can instead start with

$ docker-machine start neuralart

When the machine is running, use the following to set environment variables appropriately:

$ eval "$(docker-machine env neuralart)"

To build a Docker image with the Dockerfile in this repository, use

$ docker build -t <your dockerhub name>/tfjupyter:<version number> .
$ docker build -t <your dockerhub name>/tfjupyter:latest .

You may push your image to DockerHub with

$ docker push <your dockerhub name>/tfjupyter:<version number>

once you have an account there. You may retrieve a pre-built image anonymously

$ docker pull gnperdue/tfjupyter

To check the IP address where the Docker machine is running:

$ docker-machine url neuralart | \
>     perl -ne '@l=split(":");print split("/",@l[1]);print "\n";'

Forward port 8888 to 5000, locally mount $PWD to the notebooks directory in the Docker container:

$ docker run -p 5000:8888 -v $PWD:/notebooks -it --rm gnperdue/tfjupyter

(Or, instead of gnperdue, use your DockerHub username.)

Now, direct your browser to the IP address returned by docker-machine earlier and set the port to 5000, e.g.

192.168.99.100:5000

To stop, use

$ docker-machine stop neuralart