In the midst of Covid19, there is a spike in demand for machine learning solutions that can help tackle the pandemic or ease the societal transformation back into a sense of normality. Given that many countries are now easing lockdown restrictions, products that can assist in the controlled reopening of public functions whilst limiting the spread of the virus are particularly valuable. One way to do so is in the use of face masks which is increasingly being implemented across the world, with some countries and travel operators making usage mandatory. Whether or not people are complying with the new rules can be challenging to assess while preserving privacy.
AI to inform policy
That’s where this new solution comes in, providing a tool to tackle the issue. Utilising multiple camera streams, the Deep Learning-powered face mask detection system can help regulatory functions monitor if people are wearing a mask or not in real-time. The system is able to do so by first detecting and extracting all faces identified in a video stream using a neural network. Next, the system crops into the identified faces to perform another scan, this time to specifically look for a mask using a Resnext visual AI architecture. From this not only will it provide a yes/no response to the question “is this person wearing a mask?” but it will simultaneously do so for all the persons identified in the video stream. So if you are monitoring a crowded public space, this new tool will provide a live percentage estimate of compliance with face mask recommendations. Importantly, the system does not record any footage nor perform any kind of personal identification, thereby guaranteeing privacy which is always a must when it comes to visual AI systems.
The tool’s potential use is widespread. From supermarkets, cinemas to airports, airlines and government offices, the product can assist in ensuring the rules are followed where required and therefore help inform public policy in real-time, a valuable contribution when facing the fast-moving Covid-19 pandemic.
Zindi and InstaDeep – an example of productive collaboration
The tool originated in April, when InstaDeep’s AI Engineer Mohamed Jedidi, who had recently placed second in an unrelated international computer vision competition, collaborated with Zindi to prepare the data for Zindi’s ‘Spot the Mask’ computer vision challenge. The challenge was successful, enlisting 425 data scientists who collectively produced close to 150 solution submissions in just 60 hours, showing again the power of the data science community to quickly deliver technical solutions, in this case over the course of a single weekend.
The work, however, did not stop there. The competition focused on purely classifying on a single face whether a mask is worn or not. A team at InstaDeep led by Mohamed Jedidi built on this and quickly managed to add multiple face detections and counting in real-time. They grew the training dataset from one thousand to forty thousand images, making it possible to significantly increase accuracy.
Free usage for all
The collaboration between InstaDeep and Zindi, which yielded a production-ready AI tool at unprecedented speed, is a clear example of what can happen when talented data scientists meet on the right platform and unite their efforts. Keeping up with this collaborative spirit, InstaDeep and Zindi will offer this tool for free to all parties interested. If you would like to use our face detection tool free of charge please get in touch via hello@instadeep.com.
