Facial recognition firms are scraping photos from Instagram to train AI to see through face masks
Your selfies with face masks. Your selfies without face masks. Researchers are prowling the internet for publicly available photos and frantically trying to train facial recognition AI to see the whole picture.
VICTOR HABBICK VISIONS/SCIENCE PHOTO LIBRARY/Science Photo Library/Getty Images
The total value of the facial recognition market by 2024, according to recent research from MarketsandMarkets.
This recent market forecast was published in the summer of 2019 — well before COVID-19 hit. With the pandemic ensnaring the world, face masks have become ubiquitous — and a challenge for algorithm training.
Engineers and scientists have begun to train AI with publicly available photo sets (Instagram, for instance) so that it might be able to decipher partially covered faces. In April, an endeavor by the AI startup Workaround went up on Github. Researchers call it the COVID19 Mask Image Dataset.
The number of photos Workaround researchers used to identify selfies with medical masks, non-medical masks, and no masks.
Prediction accuracy exhibited by Workaround's facial recognition system.
“We were inspired by all the companies that were launching free tools and everything they can do to help. We have these public images from Instagram, so these are not private images. We were just searching and getting the right data.”
Yet despite the existing criticism — and now struggles caused by COVID-19 — the facial recognition industry is already seeking ways to adapt to and navigate a heavily masked world. To them, these are strange times no doubt — but herein lies a great marketing opportunity. Your worries about bias, invasive practices, misidentification, and more be damned.