Docker container

[jelmervdl]

Automatically building & pushing a docker container should make installation a lot easier.

pip wheels would also be nice for HPC + conda environments I suppose. But docker containers are easier for me, hehe, and would have less dependencies initially. I'll probably make a big one that also includes the dependencies for the filters.

Question though: some of the filters are cuda/pytorch based. Should the container be able to run those? I think nvidia-docker is just a superset of docker so images that would work with those would also work on normal docker, but then just run CPU only?

• Two-stage build (definitely don't ship node + npm + all those packages in final image)
• Ideally have it do something sensible like docker run -v $(pwd):/root -p 127.0.0.1:8000:8000 --rm server and docker run -ci -v $(pwd):/root --rm filter filter.yaml