jhennies/supervoxel_merging

Supervoxel merging pipeline

Semi-automatic segmentation workflow for Volume SEM datasets

Installation

From environment files

Paintera:

conda env create -f paintera-env.yml

Main environment:

conda env create -f paintera-mc-workflow-env.yml

Manually

Paintera

TODO

Main environment:

conda create -n paintera-mc-workflow-env -c cpape -c conda-forge elf python=3.7
conda activate paintera-mc-workflow-env
conda install -c conda-forge napari
conda install -c cpape z5py

Usage

from paintera_multicut_workflow import pm_workflow

pm_workflow(
    results_folder='path/to/results/folder/',
    raw_filepath='path/to/raw_data.h5,
    mem_pred_filepath='path/to/mem_pred.h5',
    supervoxel_filepath='path/to/supervoxels.h5',
    mem_pred_channel=2,  # Required if mem_pred.ndim == 4 to select channel
    auto_crop_center=True,  # Crops to the center if raw, mem or sv shapes are > annotation_shape
    annotation_shape=(256, 256, 256),
    paintera_env_name='paintera_env_new',  # Name of the paintera environment
    activation_command='source /home/hennies/miniconda3/bin/activate', 
    export_binary=True,
    conncomp_on_paintera_export=True,  
    verbose=True
)

Installation and usage from archive (under development)

Linux

Download pm_workflow.tar.gz

Unpack with

mkdir pm_workflow
tar -xzf pm_workflow.tar.gz -C pm_workflow

Run the pipeline

cd pm_workflow
./run_workflow.sh input_folder result_folder [arguments]

Use help for description of arguments

./run_workflow.sh -h

Windows

Download pm_workflow_win.zip

Unpack to the the disired location (workflow directory)

Run the pipeline using the command prompt:

Open command prompt and navigate to the workflow directory, e.g.

cd path\to\pm_workflow_win

Run the pipeline with

run_workflow.bat input_folder result_folder [arguments]

Use help for description of arguments

run_workflow.bat -h

supervoxel_merging's Issues

Don't provide fully frozen conda envs

Specifying fully frozen conda envs as default way of installation is not a good idea.
This has two major drawbacks:

• It will not be very portable to other operating systems or python versions, because pacakges are often not available in the exact same pinning.
• The env file will become invalid as soon as one of the packages is removed from a conda-forge channel you depend on.

For example, I cannot install paintera-mc-workflow-env, because it fails with this error:

Collecting package metadata (repodata.json): done
Solving environment: failed

ResolvePackageNotFound: 
  - nifty==v1.0.7=py37_3