Challenge Website

1. Requirements

2. Dependencies and installation

3. Usages

Tested with:

• Ubuntu 18.04/20.04 | Python 3.8-dev | CUDA 10.1

GPU memory requirements:

• Default training pipeline requires _ GB memory

• Default inference pipeline requires _ GB memory

Pytorch

Follow Pytroch instructions for installation. Run the following code to verify:

python -c 'import torch; print(torch.rand(4, 2, device="cuda"))'

Monai - more info at https://docs.monai.io/en/latest/installation.html

pip install monai

Nibabel - more info at https://nipy.org/nibabel/

pip install nibabel

DiskCache - more info at https://pypi.org/project/diskcache/

pip install diskcache

Learning Rate Finder - more info at https://github.com/davidtvs/pytorch-lr-finder

pip install torch-lr-finder

Fast Progress - more info at https://github.com/fastai/fastprogress

pip install fastprogress

Clone this repo and follow instructions in run_net_clean.ipynb. This requires installation of jupter notebook or jupyter lab. Check run_net.ipynb to see what the output should look like.

Generate a metadata file and prepare a cache of preprocessed data. This will create a metadata/df_meta.fth file as well as a cache/ folder in the current directory.

python prepcache.py --data-path "COVID-19-20_v2/Train"

Use lr finder to find a good learning rate. This will run the model through a hundered batches which take a few minutes to complete on the gpu.

python modules.util.lr_finder.py

Begin training. Validation will run every 5 epochs and a copy of the model parameters will be saved under saved-models/ folder. The best model based on F1 score is saved with the prefix .best.state

python training.py --lr 0.001

Prepare the metadata and cache files. Note that this will overwrite the metadatafile generated from the train set if it was created previously

python prepcache.py --data-path "COVID-19-20_TestSet"

You can specify a UID from a CT scan based on the number found in the filename. For example, pass in 0180_0 for volume-covid19-A-0180_0_ct.nii.gz

python inference.py '0180_0' --data-path "COVID-19-20_v2/Train"

To run inference on all of the files, simply pass the --run-all flag as shown below

python inference.py --data-path "COVID-19-20_v2/Train" --run-all