Link to method description: https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/discussion/194145
Software Requirement: torch 1.3.1 torchvision 0.4.2 pydicom 2.0.0 pandas 1.0.3 numpy 1.17.2 gdcm 2.8.9 apex 0.1 huggingface transformers 2.11.0 imgaug
Following libraries are used in my code but not required to install: https://github.com/Cadene/pretrained-models.pytorch https://github.com/lukemelas/EfficientNet-PyTorch https://github.com/albumentations-team/albumentations
Hardware Requirement: I used 4 x RTX Titan for image-level model training, and 1 x 2070 for study-level training.
For local validation, go to trainval/ and sh run.sh
To retrain on the full training data, go to trainall/ and sh run.sh