RunCellpose via docker

[callum-jpg]

Here's the RunCellpose model that's been adapted to work with docker.

This version of the module allows users to use Cellpose using the .exe/.app downloaded from the CellProfiler website. All a user would have to do is download and install docker, followed by downloading the cellprofiler/runcellpose_no_pretrained or cellprofiler/runcellpose_with_pretrained docker images and setting the plugins directory in CellProfiler > Preferences. These different docker images either contain or don't contain all of the Cellpose pretrained models

Outstanding questions:

• As of writing, the docker images cellprofiler/runcellpose_no_pretrained:0.1 and cellprofiler/runcellpose_with_pretrained:0.1 are built on the pytorch CUDA docker image and as a result are pretty beefy (12GB on disk for me). Since some users may not have a CUDA GPU, it might make sense to create a slim CPU version and allow users to select this from the GUI dropdown.
• I've not tested running this Cellpose docker on a CUDA GPU, though I'm hoping the docker flag --gpus all handles this
• Will determining docker_path work across different machines? How consistent is docker path? What is the docker path on linux?
• As the module is upgraded over time, should the module variable_revision_number be linked to a particular version of a docker image? Hopefully this would ensure compatibility
• I think how the custom model is loaded should be changed: a user should just have to find the model file itself, rather than finding the directory in which it's contained and then manually typing the model file name

[bethac07]

Awesome! One comment from me

As the module is upgraded over time, should the module variable_revision_number be linked to a particular version of a docker image? Hopefully this would ensure compatibility

I don't think so, Docker versioning SHOULD take care of that.