University of Washington researchers have been testing facial recognition systems on a massive scale over the past several months, and the results are starting to come in.
Called the MegaFace Challenge, the academic institution’s initiative is a public competition that tests facial recognition systems against databases reaching a million individuals. Some algorithms that had performed well with smaller sample sizes (as in, several thousands of people) saw debilitating drops in accuracy, while others managed to maintain relatively impressive performance levels; all suffered, however.
Google’s FaceNet system was one of the strongest performers, dropping from near-perfect accuracy to about 75 percent in one test; while Russia’s N-TechLab technology dropped to 73 percent. Meanwhile, Facebook’s DeepFace technology wasn’t submitted for the contest, so there’s no telling how its performance would compare.
For some, the results will highlight how immature facial recognition technology still is, though others may argue that they demonstrate the rapid technological advances that have been seen in the technology over the last several years. And with private and public sector interests continuing to refine their technologies, future results are likely to improve along with the state of the art.
June 27, 2016 – by Alex Perala