Download the models from the links in the pretrained models section.
Due to limited cloud space, we cannot hold the entire dataset, so we provided dataset generation scripts. To reproduce the results, we provided txt files, which contain the exact images we used for training and testing.
Possible ways to contribute: provide more data, experiment with other model prototypes, validate model on external data, medical expertise.
To create a more balanced test set, since there are only 31 covid test samples, we randomly sampled 100 from normal and pneumonia.
Since this model is designed by GenSynth and have different microarchitecture designs in each module, we do not have the code available, but we provided the training script for retraining on the given pretrained models.
GSInquire is an explainability tool that highlights regions of interest associated with predictions. The image in README was produced with it. It is a GUI tool. It is currently available to medical practitioners to trial it as a way of assisting making informed decisions. From README: If you are a researcher or healthcare worker and you would like access to the GSInquire tool to use to interpret COVID-Net results on your data or existing data, please reach out to [email protected] or [email protected]
The models are not in the repo since they are large and over the Github file size limit of 100 MB. You have to download the models first and change the paths to point to where you put the models. Refer to steps in the Training and Evaluation section of the readme.