Face mask selfies are being used to retrain facial recognition systems

Shawn Knight

Posts: 12,193   +120
Staff member

Millions of people started wearing face masks in public settings after the Centers for Disease Control and Prevention (CDC) recommended doing so last month to help slow the spread of Covid-19 when social distancing measures are difficult to maintain.

In our social media-driven world, it was inevitable that selfies and photos of people donning masks would soon flood popular platforms like Instagram and Facebook. That’s exactly what has happened and it’s apparently become quite the issue for facial recognition platforms.

As CNET correctly highlights, masks cover a significant portion of your face when worn correctly including the nose, mouth and chin. With these valuable identifiers suddenly out of the equation, facial recognition companies are scrambling to retrain their algorithms to lean more heavily on other attributes and they’re reportedly using photos sourced from public social media accounts to do so.

CNET said it found thousands of “face-masked selfies” available in public data sets. This batch on GitHub, for example, contains more than 1,200 samples sorted by people wearing medical masks, no masks, or non-medical (DIY) masks. The publication also cited a collection of more than 5,000 photos of masked individuals sourced from China a month earlier.

The group behind the 1,200 sample set, AI startup Workaround, said they collected the data from public Instagram accounts. CEO Wafaa Arbash told CNET they found the images by searching hashtags related to masks. The users weren’t asked for permission to be included in the database.

A spokesperson for Facebook, which owns Instagram, told the publication they do not allow third parties to collect or use photos posted by their users in this way without their consent. “We are continuing to investigate this,” the spokesperson added.

Masthead credit: wanderingstock, Vitaly Art

Permalink to story.

 

p51d007

Posts: 2,366   +1,626
Look at it this way. Get everyone to wear a mask, tell them it is for safety. Once you get a majority of the population wearing a mask, retrain the AI in facial recognition to figure out who it is under the mask. Then IF the world governments deem it "safe" for people to not use a mask, then it will be easier for them to figure out who is wearing a mask to hide their face, say when they rob something, or are at a protest trying to hide their face.
Hey, if I can think of it, you know the "we have to know everything" bunch can.
 
  • Like
Reactions: Theinsanegamer

mbrowne5061

Posts: 1,459   +803
Look at it this way. Get everyone to wear a mask, tell them it is for safety. Once you get a majority of the population wearing a mask, retrain the AI in facial recognition to figure out who it is under the mask. Then IF the world governments deem it "safe" for people to not use a mask, then it will be easier for them to figure out who is wearing a mask to hide their face, say when they rob something, or are at a protest trying to hide their face.
Hey, if I can think of it, you know the "we have to know everything" bunch can.
They can already figure out who you are with other visual biometrics. Walking gait is a big one that is pretty much impossible to fool - even with a limp - and it works both from a distance and with the lower-res footage that most CCTV systems use (unlike facial recognition). Facial recognition is more used to sell you stuff than it is used by governments to track you.