The “wild” face dataset corresponds with the real-world use of face recognition.
The latest results from the National Institute of Standards and Technology (NIST) show that Innovatrics face recognition algorithms are among the best in the world. In fact, in the wild face dataset, the Innovatrics algorithm emerged as the most accurate in the world, besting over 100 submissions.
FRVT Verification as of 04/15/2019: False non-match rate (FNMR) on Wild Faces (evaluated for FMR (False Match Rate) = 0.005)
The dataset consists of faces extracted from real-world footage, often made in subpar lighting conditions in different positions and partially obscured by hair, hands, or objects. This is what the algorithm should be able to deal with when searching for specific faces in a security camera footage.
“Our long-term focus is on leveraging AI technology to identify and verify real-world faces as accurately as possible. This result is a confirmation of our team’s efforts to continuously improve and deliver the best face recognition solutions on the market,” explains Marián Beszédeš, Head of Research and Image Processing at Innovatrics.
Face Recognition Vendor Test (FRVT) is an ongoing test of over 100 algorithms submitted by biometric companies from all over the globe. It checks the error rate in verification of different types of images, specifically the inability to recognize a given face on an image.
The full report can be found here (PDF).