Researchers scrape social media for masked faces to train AI models


After governments in many parts of the world began mandating wearing masks when out in public, researchers in China and the US published datasets of images of masked faces scraped from social media sites to use as training data for AI facial recognition models. Researchers from the startup Workaround, who published the COVID19 Mask image Dataset to Github in April 2020 claimed the images were not private because they were posted on Instagram and therefore permission from the posters was not required. Instagram's owner, Facebook, responded that third parties are not allowed to collect or reuse posted photos without consent.

Writer: Alfred Ng
Publication: CNet