CyberLink climbs NIST FRVT rankings for facial recognition accuracy in wild category
CyberLink has announced that its FaceMe AI Facial Recognition Engine has scored a biometric accuracy rate of 97.02 percent in the WILD 1E-4 category of the latest Facial Recognition Vendor Test (FRVT) from NIST, ranking it behind algorithms from only 11 other test participants.
The FRVT 1:1 results released July 3 show results with a dataset of faces from surveillance camera footage or photos, including a wide range of real-world situations, such as capture angles, poor lighting, or partially covered faces.
The WILD 1E-4 category reflects use cases such as smart retail environments, smart surveillance, and smart city deployments, according to the announcement. The company says its showing in the top tier of participants demonstrates FaceMe’s reliability and precision for implementation in a range of IoT and AIoT edge devices. FaceMe is currently being used for retail trials in Asia, and CyberLink is planning to launch U.S. trials soon.
“Since its release, FaceMe has achieved outstanding results in major facial recognition competitions, including NIST and the MegaFace Challenge. The results confirm FaceMe‘s global leadership in the facial recognition industry, providing world-class reliability, accuracy, and performance,” says CyberLink CEO Dr. Jau Huang. “The WILD 1E-4 test reflects real-world usage scenarios. The results of this test allow system integrators and developers to identify the most accurate and reliable facial recognition technology to implement in smart surveillance and smart retail scenarios.”
FaceMe also ranked 12th in the MegaFace Challenge with an accuracy rate of 98.41 percent.