nseg - Comparing Methods for Neuron Instance Segmentation

Code for experimentation with neuron instance segmentation methods based on multi-task learning and Local Shape Descriptors.

Installation

Run from the repository clone directory:

conda env create -f environment.yml
conda activate nseg
pip install .

Configuration

All configuration files are in the nseg/conf directory. To make config changes you can either change the files themselves or pass config paths or option overrides through the [Hydra] (https://hydra.cc/docs/advanced/hydra-command-line-flags/) CLI of the entry point scripts.

Data preparation (j0126)

Run in this order:

python -m nseg.scripts.rename_gt
python -m nseg.scripts.zarrify_j0126_gt
python -m nseg.scripts.create_data_split

Training

nseg-train

Small-scale inference and evaluation (small cubes)

nseg-segment

Large-scale inference and evaluation (whole dataset or ROI)

nseg-eval i1_predict.model_path=<PATH_TO_TRAINED_MODEL>