In this project we created a transfer learning based COVID-19 lung x-ray classifier WebApp for the Scitek - POATEK Datathon. We won based on technical rigor :)
In order to run it, just follow these instructions:
- Clone the repository with
git clone https://github.com/gfc-fiscomp/xcovid.git. - Get into the repository directory with
cd xcovid. - Type
streamlit run app.pyon the terminal. A localhost instance should then open up in your browser! (note: the website is entirely in Brazilian Portuguese)
The model used for image classification is a ResNet50 pre-trained with ImageNet weights that was finetuned using the images providade in this dataset (given to us by the datathon organizers). In order to learn more about how the model was trained, check out the rede.ipynb file over here. We had to deal with a very imbalanced dataset! The communication between the WebApp and the model is made possible by call_model.py, which can be seen over here.
Here is the classifier's confusion matrix on test data:
If you'd like to test the WebApp for yourself we have two example images available in the example_images folder.
Home page:
Test page:
A classified image:
About page:
