We need to wear masks to stop the spread of COVID-19, as a first layer of defense. However, many forget to wear them or do not pay heed to guidance. Therefore, Masky aims to use awareness and peer pressure to convince people to wear a mask. No one wants to be the odd man out.
What it does
Masky is a tool that highlights faces of people that are not wearing masks. These faces can be optionally blurred.
How I built it
Masky was built with a variety of technologies, consisting mainly of a frontend and a backend. I wanted to create a web interface so that it would be usable on any device.
The model that Masky uses is Yolov5s. The smallest and quickest version of Yolov5, a recently released state-of-the-art object detection model. Currently, requests to the server happen every second and CPU-only processing takes approximately 100ms.
The model was trained in Google Colab.
Challenges I ran into
There were quite a few issues in the training process, making it take longer than expected. Additionally, the regular inference script was not designed to take an input of a base64 string and was not designed to work without command line parameters. Those changes took considerable work.
Accomplishments that I'm proud of
What I learned
How to create a full fledged application in a short period of time.
What's next for Masky
Integration with internal security systems, on-device inference, accessible API.