neurosegmenter
creating an Anaconda env
Can use the provided yml env configurator to build a ready-to-use anaconda environment.
conda env create --file conda_env.yml
conda activate neuroseg
Building a docker container
first build the package
python setup.py bdist_wheel
then build the docker image
docker build -t neuroseg:latest .
Pushing the container to ATLANTE LENS Registry
retag the image to the registry on atlante.lens.unifi.it
docker image tag neuroseg:latest atlante.lens.unifi.it:5000/neuroseg:latest
before pushing you need to be able to access the regisry, we can make a tunnel to atlante bouncing on liquid
sshuttle -r castelli@atlante.lens.unifi.it -x liquid.lens.unifi.it 150.217.0.0/16 150.217.157.89
and then we push
docker push atlante.lens.unifi.it:5000/neuroseg:latest
Running an example script
sudo docker run -it -v /home/castelli/neuroseg/examples:/opt/examples neuroseg:latest python /opt/examples/example_script.py
resolving TF 2.7.0 issues: wrong CUDA libs
If TF 2.7 doesn't find the correct CUDA libs in an anaconda env it might depend on how libraries are loaded by TF after 2.7
just set LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/home/phil/anaconda3/envs/neuroseg/lib