3rd of Task3 and 5th of Task4 of StructSeg2019 competition - MICCAI 2019

This repository is the solution of 3rd place of Task3 and 5th place of Task4 of StructSeg2019 competition which is a part of MICCAI 2019.

• catalyst==19.9.1
• albumentations==0.3.2
• segmentation-models-pytorch==0.0.2

You may see my model named as VNet, you may be confused to this paper. I named as V because of my personal purpose (The full name is VUONGNet).

Change the input and output path in extract_slices.sh.

bash bin/extract_slices.sh

The output should contain numpy array of each slice and a csv file (data.csv)

python src/preprocessing.py split-kfold --csv_file <path to your data.csv file> --n_folds 5 --out_dir <path to your output> 

I train 2D images for 2 tasks. All the settings are placed at: config_2d.yml

bash bin/train_2d.sh