FLANNEL

Data Prepare

Data Collect

1. Download CCX data: from https://github.com/ieee8023/covid-chestxray-dataset, put them into original_data/covid-chestxray-dataset-master
2. Download KCX data from https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia, put them into original_data/chest_xray

   Data Preprocess

3. extract data from CCX: data_preprocess/get_covid_data_dict.py
4. extract data from KCX: data_preprocess/get_kaggle_data_dict.py
5. reorganize CCX&KCX data to generate 5-folder cross-validation expdata: data_preprocess/extract_exp_data_crossentropy.py

Model Training

Base-modeler Learning

FLANNEL/ensemble_step1.py for 5 base-modeler learning [InceptionV3, Vgg19_bn, ResNeXt101, Resnet152, Densenet161]

(E.g. python ensemble_step1.py --arch InceptionV3)

ensemble-model Learning

FLANNEL/ensemble_step2_ensemble_learning.py

Codes for

@misc{

  title={FLANNEL: Focal Loss Based Neural Network Ensemble for COVID-19 Detection}, 

  author={Zhi Qiao and Austin Bae and Lucas M. Glass and Cao Xiao and Jimeng Sun}, 

  year={2020}, 

  eprint={2010.16039}, 

  archivePrefix={arXiv}, 

  primaryClass={eess.IV} 

}