NIST FRVT

The US government's independent benchmark for face-recognition accuracy.

It works like a public report card. Companies submit their face recognition software, a government lab tests it on its own data, and the results are published for anyone to compare.

The National Institute of Standards and Technology (NIST) Face Recognition Vendor Test, or FRVT, was a continuously published evaluation of algorithm accuracy and demographic differentials. NIST has since rebranded and split it into two activities: FRTE, the Face Recognition Technology Evaluation, and FATE, the Face Analysis Technology Evaluation.

The two activities divide the work this way:

• FRTE (Face Recognition Technology Evaluation) covers tracks that pertain to identity verification, including one-to-one matching, one-to-many identification, demographic effects, face masks, and twins.
• FATE (Face Analysis Technology Evaluation) covers tracks that process and analyze face images, including morph detection, image quality, and age estimation and verification.

NIST describes both as independent government evaluations. The programs run continuously, and developers can send new submissions on a regular schedule.

The NIST FRVT sits next to terms that are easy to mix up:

• Face match is the one-to-one comparison that FRTE tests.
• Equal error rate is one of the measures commonly used to compare algorithms.
• Demographic differentials are the variations in error rates across groups that NIST measures and reports.

For more, see Incode's Stellar Performance in NIST FRTE 1:N Identification Testing.