/Kaggle-UWMGIT

code for kaggle: UW-Madison GI Tract Image Segmentation

Primary LanguagePythonApache License 2.0Apache-2.0

GUIDE

  1. download data from kaggle and unzip all to ./data/tract

  2. run ./data/tract/data.ipynb under ./data/tract to convert directory to mmsegmentation format (run mkdir if necessary)

  3. run ./tools/dist_train.sh ./work_configs/tract/baseline.py $NGPUS, $NGPUS depends on your own devices

  4. make a kaggle submission notebook refering ./data/tract/submission.ipynb

UPDATE

Competition solution and submission code have been open sourced at Kaggle

aAll config files have been uploaded to ./data/tract/final_solution, saved in classification_configs and segmentation_configs respectively. These configs are for referencing only. To reproduce, there is much to do with data, including data cleaning and preprocessing, so I have no plan to include this part.

3D monai codes are saved at ./monai, with only slight changes compared with public monai codes.