This repository provides a PyTorch implemenation of the classifcation problem in the Kaggle SIIM-ISIC Melanoma 2020 Competition, https://www.kaggle.com/competitions/siim-isic-melanoma-classification/overview, (It can be modified for other problems though).
Using an ensemble of models produced by this code as well as other TensorFlow models, our team managed to be in top 10% and win a bronze medal in the competition.
The implementation uses a 5-fold cross-validation method.
To train a model, specify configurations in config_file.py and run :
python train.py --save_folder config55 --dev_num 0
where save_folder will be the name of the created folders to save best checkpoints and predictions.
(Some updates might be introduced later).