Lung-Instance-Segmentation-Workflow

(Instance segmentation with U-Net/Mask R-CNN workflow using Keras & Ray Tune)

Lung instance segmentation workflow uses Chest X-ray for predicting lung masks from the images using U-Net model.

Running the Workflow

1. Clone the repo: git clone https://github.com/pegasus-isi/lung-instance-segmentation-workflow.git
2. Run using the sample dataset: python3 workflow.py --lung-img-dir inputs/train_images --lung-mask-img-dir inputs/train_masks

To Run the workflow using the production dataset, you must first obtain it.

1. Install the Kaggle Python package: pip3 install kaggle
2. Download the dataset: python3 get-dataset.py
   • The dataset is also backed up at /lizard/projects/ml-workflows/lung-segmentation-workflow/lung-segmentation-data.tar.gz and has a sha256 hash of f30a3d450dce65a4c0f93c9e408e0cd457023d70db9599032179c36758fbf5fc
3. Run python3 workflow.py
   • by default the script will look for data in ./data which was created by get-dataset.py

Executing Standalone Scripts

1. Use the command pip3 -r requirements.txt to install the required packages
2. Go back to the lung-instance-segmentation-workflow directory and make a directory called output
3. Download the dataset by running the python script called "get-dataset.py" by python get-datatset.py
4. Execute the end-to-end.sh script