This is basically set up for heroku-style deployment, although you don't have to (and probably shouldn't) use heroku for this. The built docker image is like 2.5gb so be careful of exceeding disk space. Probably a good idea to set resource limits as well otherwise people can dos you with big images. (Nginx in front of the app server may reject requests that are too large, but don't count on it)
Due to the use of cpu-only pytorch the requirements.txt is tied to python3.8/linux/x86_64. If you want to run on some other combination of environment you need to tweak those requirements, you can find the links to pytorch wheels here.
If you want to run locally just type gunicorn app:app after installing dependencies.
automatically resize frames to multiples of 64automatically convert frames to 8bpc RGB removing alpha if it's thereautomatically downsample frames to something reasonableShow on image the flow and weight map- More detailed error messages