US facial recognition developers jockey for position in NIST accuracy testing
U.S. developers do not appear among the first seven algorithms of the leaderboard for the Face Recognition Vendor Test for verification posted by the U.S. National Institute of Standards and Technology on June 16. Only 2 appear among the top 20, overall.
A handful of vendors based in America show up among the leaders in face biometrics accuracy, however, in the NIST FRVT 1:1 Verification, and multiple startups have made impressive showings.
An algorithm submitted on May 10 by Omnigarde, a startup based in Southern California, is near the top of the third page out of 18 on the NIST leaderboard, scoring accuracy among the top 50 entries from all vendors worldwide in three different categories.
The “omnigarde-003” algorithm had a 16 percent lower error rate than Omnigarde’s previous entry, according to a company announcement.
Omnigarde says the memory required, feature vector size, feature extraction time, and matching time of its facial recognition are among the best of the most accurate algorithms analyzed by NIST.
An algorithm submitted at the beginning of 2023 by Atlanta, Georgia-based startup Armatura cracked the top-20 in the VisaBorder Yaw≥45 degrees and Border categories in the June 16 update.
The top overall result by a U.S. vendor is an algorithm from Paravision, which was found to have the best accuracy with the new VisaBorder Yaw≥45 degrees dataset.
Rank One Computing’s latest entry, submitted in late-2022, was among the 15 most-accurate algorithms matching three datasets. The previous submission from ROC also appears among the top 50 algorithms for overall accuracy among the more than 500 evaluated by NIST.