/SIIM-ACR-Pneumothorax-Segmentation-5th

https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/leaderboard

Primary LanguagePython

SIIM-ACR-Pneumothorax-Segmentation-5th

https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/leaderboard

Dependencies

  • python==3.6
  • torch==1.0+
  • torchvision==0.3+

Solution

https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/discussion/107603#latest-620540

prepare data

competition dataset

./input/dicom-images-train
./input/dicom-images-test

NIH data

download NIH dataset
https://www.kaggle.com/nih-chest-xrays/data
common.py: EXDATAPATH = './input/NIH/images/*.png'

Train

python train_semi.py --modelname="seresnext50" --fold_index=0 --batch_size=16 --lr=3e-4

Test

stage1: python test_json.py
stage2: python test_json_stage2.py

ensemble

stage1: python ensemble_forsub1.py
stage2: python ensemble_forsub1_stage2.py