Skip to content

Latest commit

 

History

History
26 lines (14 loc) · 1.44 KB

README.md

File metadata and controls

26 lines (14 loc) · 1.44 KB

Runmila-AI-Institute-minoHealth-AI-Labs-Tuberculosis-Classification-via-X-Rays-Challenge

competition website

  • Tools: Google Colab, Pytorch, Albumentation, Pytorch pretrained models, Resnet152

  • Techniques: Transfer Learning, Data Augmentation

  • See TB.ipynb for package installation details, stored in notebook folder

  • The data used is the small version: train_small.zip/ test_small.zip stored in the inputs folder

  • Training: 5 folds cross-validation , however, fold 2 is the one that achieved the highest submitted accuracy

  • TB_stratified_kfold.csv is the 5 folds cross-validation file stored in the inputs folder, you can generate it using the train_fold.py file stored in the folder

  • Submission results are stored in nsub.TB.csv in the inputs folder

  • Training engine is in utils.py stored in source folder alongside with train_folds.py

  • The training models are saved in the models folder

  • To run the code please store your data according to the folders as disccused above and don't forget to change paths in the main notebook.

  • You can add inputs folder to store your data and another folder called models to save your best trained models

  • My Zindi Profile.