CyberLink is once again celebrating a strong performance in the NIST’s Facial Recognition Vendor Test (FRVT). The company’s FaceMe algorithm achieved a top-six score on the latest one-to-many WILD evaluation, and was the most accurate solution developed outside of China.
The NIST’s WILD (1E-5) test is designed to assess the performance of facial recognition algorithms in challenging real world scenarios. FaceMe was able to identify faces with 96.98 percent accuracy, even when those faces were captured at a steep angle, in low light, or were partially obscured in some fashion.
In addition to its strong overall score, FaceMe received high marks for its mask detection capabilities. The solution was the seventh best outside of China and Russia, which corresponds to a 50 percent improvement over its previous score. The company updated its platform to enable mask recognition in the early days of COVID-19, and later supplemented that with temperature detection to help combat the spread of the disease.
CyberLink has previously claimed that FaceMe is 95 percent accurate when asked to identify people wearing masks. However, its latest test results indicate that the company has managed to refine and improve its technology in the past few months.
“The WILD test is one of the most competitive tests in the industry, and we’re thrilled about FaceMe’s latest ranking,” said CyberLink CEO Jau Huang. “We are committed to providing the best-in-class, most accurate and most secure solution to the market and will continue to innovate and improve the FaceMe algorithm to best suit end-user needs.”
CyberLink noted that many Chinese and Russian facial recognition solutions are not commercially available in other parts of the world. The company has scored well on the FRVT in the past, and has made considerable gains over the course of the past year. FaceMe posted a top-18 score in the spring, and followed that up with a ninth-place finish in December.
January 29, 2021 – by Eric Weiss