8th Solution of SIIM-FISABIO-RSNA COVID19 Detection Competition

Environment Setup

Anaconda virtual env is recommended. Python version is 3.7.

pip install -r requirements.txt

Dataset Preparation

pip install kaggle # Kaggle API
vi ~/.kaggle/kaggle.json # Kaggle Profile - Account Tab - API - Create New API Token  ex) {"usernames":"jihunlorenzopark", "key": "xxxxx"}

mkdir data-640 # Naming format is data-{image_size}
cd data-640
kaggle datasets download -d jihunlorenzopark/siim-fisabio-rsna-data-640 # Generated from the code here https://www.kaggle.com/c/siim-covid19-detection/discussion/239918
unzip siim-fisabio-rsna-data-640.zip

# For NIH CHEST X-ray dataset pretraining, run the below
mkdir data-nih
cd data-nih
kaggle datasets download -d jihunlorenzopark/nih640
unzip nih640.zip

Run

Training script is run.py configured by config.yaml. Also, there are some model specific parameters in unet_smp.yaml.

Before running the script, update root parameter in config.yaml to your cloned directory.

Also, unless you use wandb for logging training metrics/losses, turn off logger option to use default build-in logger of pytorch. i.e. logger=False

# Please refer commands in this experiment.bash file.
bash experiment.bash