SalDACov: Improving Data Augmentation applied to the COVID-19 lung CT Segmentation with a novel technique based on Visual Salience

See the implementation and the available networks in: Segmentation Models

Installation:

pip install -r requirements.txt

Copy the files inside the scripts to the utils folders inside the lib/python3.6/site-packages/segmentation_models_pytorch folder

Dataset:

Setup a txt file with the images paths as follows for training and validation:

path/to/image1.jpg path/to/mask1.png
path/to/image2.jpg path/to/mask2.png
path/to/image3.jpg path/to/mask3.png

or just the images paths for tests

path/to/image1.jpg
path/to/image2.jpg
path/to/image3.jpg

Setup the config file following the examples in the config folder

Execute:

python main.py --configs config_file.yml

Images Generation:

StyleGAN ADA Pytorch
StarGANv2

Setup a images as following:

images/images/*.jpg (images from the dataset)

masks/lesion_masks/*.png (masks with lesions from the dataset)

lungs/lung_masks/*.png (predicted lung masks)

gan/images/*.jpg (images generated by the GAN

gan/predicted_masks/*.png (predicted lung masks)

Execute:

python augmentation.py gan

Citation

@article{dacov2023,
    author = {Bruno A. Krinski and Daniel V. Ruiz and Rayson Laroca and Eduardo Todt},
    title = {DACov: a deeper analysis of data augmentation on the computed tomography segmentation problem},
    journal = {Computer Methods in Biomechanics and Biomedical Engineering: Imaging \& Visualization},
    volume = {0},
    number = {0},
    pages = {1-18},
    year  = {2023},
    publisher = {Taylor & Francis},
    doi = {10.1080/21681163.2023.2183807},
    URL = {https://doi.org/10.1080/21681163.2023.2183807},
}

SalDACov Evaluation

SalDACov with Generic Augmentation Techniques Evaluation

DACov Evaluation

Generic Augmentation Techniques Evaluation