GAN-based Data Augmentation for class imbalance in Chest X-ray

Download Data

$ bash bash_scripts/run_data.sh

Dataset	Link	Download Directory
RSNA	https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/data	data/RSNA_Pneumonia
COVID-19	https://github.com/muhammedtalo/COVID-19	data/COVID-19
COVID-chestxray-dataset	https://github.com/ieee8023/covid-chestxray-dataset	data/covid-chestxray-dataset

Baseline (Supervised DenseNet121)

$ python main.py --with_gan "" --idx 0 --user "vasu" --skip_training ""  --dataset_size  10  --dataset  "COVID"  --fraction  0.5  --epochs  30  --data_aug  "True"

For more info, read about the arguments $ python main.py --help

GANs

GAN	Path	Loss
DCGAN	scripts/GANs/DCGAN.py	BCE Loss
WGAN	scripts/GANs/DCGAN_GP_conditional.py	W-Loss + Gradient Penalty
SNGAN	scripts/GANs/SNGAN.py	Spectral Normalization
LSGAN	scripts/GANs/LSGAN.py	Least Sqaures GAN
IMPROVED BAGAN	scripts/GANs/improved_bagan/imp_bagan.py	Gradient Penalty

$ python scripts/GANs/SNGAN.py --dataset COVID-small --user vasu --with_gan True --epochs 200 --lr 2e-3 --display_step 100 --z_dim 128 --GAN_type SNGAN --batch_size 32  --im_channel 3 --patience 30 --n_class_generate 0 --num_images_per_class 250

To run Improved BAGAN: $ python scripts/GANs/improved_bagan/imp_bagan.py

For more info, read about the arguments $ python scripts/GANs/SNGAN.py --help

Acknowledgements

This repository makes use of the code from the following repositories. We thank all the authors for making their code publically available.

• https://github.com/ssundaram21/6.819FinalProjectRAMP
• https://github.com/GH920/improved-bagan-gp