Researchers scrape masked faces from social media to train algorithms
Researchers are scraping social media posts for images of mask-covered faces to use to improve facial recognition algorithms. In April, researchers published to Github the COVID19 Mask Image Dataset, which contains more than 1,200 images taken from Instagram; in March, Wuhan researchers compiled the Real World Masked Face Dataset, a database of more than 5,000 photos of 525 people they found online. The researchers have justified the appropriation by saying images posted to Instagram are public and their use is not commercial. In some cases, facial recognition providers have asked their own staff to send in masked selfies, while NIST is digitally adding masks to photos in order to test facial recognition algorithms.
Writer: Alfred Ng