During the workshops there are a lot of interest from biologists to use NNInteractive. However, it does take a while to set it up the server, particularly if you are not familiar that kind of thing.
In MorphoCloud, we can map a read-only share to our instances. If I create the virtual environment and install the NNinteractive server there, can the users safely activate the environment and run the the command nninteractive-slicer-server --host 0.0.0.0 --port 1527 from that location; i.e., does the server need any write access to the folder while its running.
It seems to work locally. I think I will test on MorphoCloud next workshop.
Meanwhile, it would be nice if the NNINteractive can be self-contained and install the server under Slicer. That way its footprint will be a bit smaller and possibly time to install to start using will be faster as there won’t be two different copies torch and other cuda related large libraries installed in two different environment (venv for the server, and inside the Slicer for the extension itself).