AnyVision facial recognition scores among FRVT 1:1 leaders, CloudWalk revealed as MoonTime developer
AnyVision has earned top rankings for biometric accuracy across all five category groups in the latest Face Recognition Vendor Test (FRVT) 1:1 conducted by the National Institute of Standards and Technology (NIST).
The March FRVT 1:1 evaluated 177 individual biometric algorithms and reported the performance of their facial recognition technologies in one-to-one (1:1) scenarios. The test compared various face images against input images, measuring the algorithms’ accuracy, speed, storage and memory consumption, and resilience.
The ‘anyvision-005’ algorithm was not entered in the child category, but finished between 7th and 13th among all algorithms worldwide in the visa, visa border, border, wild, and both mugshot categories. The companies ahead of AnyVision on the overall 1:1 leaderboard all come from China, except VisionLabs, which is based in Russia.
“AnyVision technology is optimized for interpreting real-world video sequences on low power embedded devices, which is actually much harder to solve than what NIST focuses on, which is static images and photos, processed using Intel hardware,” explained AnyVision CTO Dieter Joecker.
NIST’s FRVT measures the performance of automated facial recognition technologies in order to assess their validity in a variety of different scenarios, including civil, law enforcement, and homeland security applications.
“The fact that we do not design our algorithms for NIST and still achieved very balanced, high rankings across all categories – including two that were not included in the initial ranking – is truly impressive,” Joecker added. “It’s a testament to how advanced our technology really is.”
The point was echoed by AnyVision CEO Avi Golan, who said that the results in the FRVT reflect the significant investments AnyVision has made to make its algorithms and AI engine among the most efficient in the market.
“We believe that AnyVision is the AI for IoT,” Golan added. “We have taken the crucial first steps towards bridging the digital-to-physical divide, with a platform for understanding the way people experience and interact with the physical world.
“In other words, we are creating the missing link in making our cities truly smart, our offices responsive, and our homes more livable,” he concluded.
AnyVision has been quite active in 2021, recently filing a patent for delivery drones with facial recognition.
CloudWalk is MoonTime and Hengrui AI
CloudWalk has been revealed as the developer behind the MoonTime and Hengrui AI Technology facial recognition algorithms, both notable for their outstanding performance in NIST’s recent FRVT tests across 1:1, 1:N, and Face Mask Effects tests.
The MoonTime ‘mt-003’ algorithm, which was first benchmarked by NIST earlier this year, tops the NIST 1:1 leaderboard after the March update, with top 5 finishes in the visa, mugshot, visa border and border categories. The same algorithm also finished second in the March update of the leaderboard for biometric accuracy with Face Mask Effects.
Hengrui’s ‘hr-000’ algorithm sits first among 1:N leaders, and places between 1st and 5th in each category except the long-duration mugshot match, which it was not tested for.