The purpose of this work is to create a WIPP plugin based on the StarDist 2D Object detection package.
src: contains the source Python codeDockerfileplugin.jsonWIPP plugin manifest
The code is written in Python 3 (tested on version 3.10) and leverages the tensorflow and stardist python packages.
conda create --name stardist python=3.10
conda install grpcio
pip install tensorflow stardist imagecodecs
conda activate stardist
docker build . -t wipp/wipp-stardist-inference-plugin:0.0.1
Pre-trained model choices from the StarDist package are: 2D_versatile_fluo, 2D_paper_dsb2018 and 2D_versatile_he
From this directory:
python ./src/stardist-inference.py \
--inputImages ./sample-data/images \
--output ./sample-data/outputs
--pretrainedModel 2D_versatile_fluo
From this directory, assuming the images to process are in a folder "sample-data/images":
docker run -v "$PWD"/sample-data:/data \
wipp/wipp-stardist-inference-plugin:0.0.1 \
--inputImages /data/images \
--output /data/outputs \
--pretrainedModel 2D_versatile_fluo
-v: mounts a volume/folder from your machine inside of the Docker container
- register the plugin.json in a deployed WIPP instance - see https://github.com/usnistgov/WIPP
- upload input images as WIPP image collection
- create a workflow by adding one step called stardist-inference
- run and monitor the workflow execution
- download resulting WIPP image colection
- WIPP team, ITL NIST, Software and System Division, Information Systems Group
- Contact email address at NIST: wipp-team@nist.gov
- StarDist Github repository: https://github.com/stardist/stardist
Uwe Schmidt, Martin Weigert, Coleman Broaddus, and Gene Myers. Cell Detection with Star-convex Polygons. International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), Granada, Spain, September 2018.