The COVID-19 Pandemic has affected us all, and has caused much confusion in Ontario, due to the various lockdowns and policy changes. Many people still remain in the dark, not knowing all of the specific policies of the pandemic and what sources to get real information from. Thus, the need for No More COVID Confusion was born, a Deep Learning Web Application that reduces the confusion about the pandemic.

What it does

No More COVID Confusion accomplishes 3 main tasks: 1) Classification of COVID-19 from CT scans 2) Recommendations for proper face masking based on Computer Vision 3) Information Center and Other features to reduce confusion and fake news about the pandemic

The user simply needs to snap a selfie of them wearing a mask to ensure that they are wearing it properly before they head outside, with 97% accuracy. If they have access to their lung CT scans, the model can also classify COVID-19 with 99% accuracy.

How we built it

No More COVID Confusion was built using PyTorch, Streamlit and Heroku. The models were trained with PyTorch Lightning Custom Training Loops. The web application was built using Streamlit, a Data Science Web Development library. Finally, it was deployed to Heroku at

Models were trained in 6 hours(overnight) on Google Cloud AI notebooks with 1 GPU accelerator and 4 Virtual CPUs.

Challenges we ran into

I ran into a variety of challenges in building No More COVID Confusion. Firstly, setting up a Google Cloud AI notebook was a challenge, due to how I had to jump through many hurdles to load the data and gain access to 1 GPU. Furthermore, deploying the model to Heroku was a big challenge, as PyTorch is massive in terms of memory. Heroku is limited to 512MB Slug Size, and thus I had to reduce my code, while still preserving model structure to lower the Slug Size of the build files.

Accomplishments that we're proud of

This was the first PyTorch model I've ever successfully deployed to Heroku, which is something I am very proud of. RUHacks gave me the opportunity to experiment with Heroku and web development more generally.

What we learned

I learned a lot about Google Cloud AI and their services in creating a Google Cloud AI notebook. Furthermore, I learned about web development and the Streamlit library, which I had never used prior to this hackathon.

What's next for No More COVID Confusion!

In the future, I plan to add further functionality to No More COVID Confusion, including a Google Cloud based Chatbot to answer any questions related to the pandemic.

Built With

Share this project: