cellpose-tools

This repository contains command line tools that allows one to run Cellpose distributed over a Dask cluster. These are used mostly by the Cellpose Nextflow pipeline.

Setup the environment

mamba env create -n cellpose-tools -f conda-env.yml
pip install -e .

Running distributed cellpose

python -m tools.main_distributed_cellpose \
    -i <input-image-or-container> -o <output-image-or-container> \
    --dask-scheduler <tcp://x.x.x.x:port> 

If no cluster is available you have 2 options:

• Use a local client that can still chunk the image into smaller blocks and pass these to a local Dask client that has a number of local workers equal to the argument specified by --local-dask-workers <nworkers>.
• Don't use any distribution and simply run Cellpose eval method on the entire image - if no --local-dask-workers is present or if --local-dask-workers 0

Image preprocessing

The tool allows for dynamic configuration of preprocessing algorithms, but so far we only support a gaussian filtering applied to the image before running the cellpose segmentation. The parameters for these preprocessing steps can be defined in a YAML file like this:

unsharp:
    sigma_one: 1.0
    weight: 0.1
    iterations: 5
    sigma_two: 0.1