kits19

Kidney Tumor Segmentation Challenge 2019

See our project report

Project structure

• ./src - source code
• ./src/starter - provided starter code from 2019 KiTS Challenge Repository
• ./data - cases downloaded from kits19/master/data
• ./data_interpolated - cases downloaded from kits19/interpolated/data
• ./report - project presentations and report
• ./*.ipynb - various research notebooks
• ./*.csv - various data statistics files
• Pipeline files:
  • ./prepare_data.py
  • ./pipeline.py
  • ./main_evaluation.py
  • ./main_train.py

To reproduce results..

Prepare data

• Download data to ./data_interpolated (Old: Download data to ./data)
• Execute data_exploration notebook -- generate data_stats.csv
• Execute h5_data_preparation notebook -- generate crops.csv and crops.hdf5 file

Prepare an environment

virtualenv --python=python3 .env
source .env/bin/activate
pip install requirements.txt

Run training

python main_train.py --checkpoint unet.pth

Technical Details

Run Tensorboard

docker run -it -p 9000:9000 -v $(pwd)/runs:/runs tensorflow/tensorflow /bin/bash
tensorboard --logdir=/runs/ --port=9000

or run this command from ./project:

docker run -d -p 9000:9000 -v $(pwd)/runs:/runs tensorflow/tensorflow /bin/bash -c "tensorboard --logdir=/runs/ --port=9000"

Port forwarding

ssh -i ssh-keys/gpu-gc -L 9000:localhost:9000 [email protected]

Use Screen

# Deattach screen
(ctrl-a-d) 
# Reattach screen
screen -r