/kaggle-rsna-intracranial-hemorrhage-detection

Primary LanguagePython

kaggle-rsna-intracranial-hemorrhage-detection

  1. Put the input files in inputs/ and unzip as the following structure:
inputs
├── stage_1_train_images
│   ├── ...
|
├── stage_x_test_images
│   ├── ...
│
├── stage_1_train.csv
│
└── stage_x_sample_submission.csv
  1. Preprocess.
python create_metadata.py
python create_windowing_images.py
  1. Train the models.
python train.py --name efficientnet-b1_102407 --arch efficientnet-b1 --apex
python train.py --name efficientnet-b2_102000 --arch efficientnet-b2 --apex
python train.py --name efficientnet-b3_102112 --arch efficientnet-b3 --apex --img_size 480 --crop_size 384
python train.py --name efficientnet-b4_102100 --arch efficientnet-b4 --apex
python train.py --name se_resnext50_32x4d_102105 --arch se_resnext50_32x4d --apex
  1. Test.
python test.py --name efficientnet-b1_102407 --hflip True
python test.py --name efficientnet-b2_102000 --hflip True
python test.py --name efficientnet-b3_102112 --hflip True
python test.py --name efficientnet-b4_102100 --hflip True
python test.py --name se_resnext50_32x4d_102105 --hflip True
  1. Ensemble.
python average.py
  1. Postprocess.
python gma.py --name average_102800