National Institute of Standards and Technology (NIST)

What is NIST?

NIST, founded in 1901 as the National Bureau of Standards, is a non-regulatory agency. In biometrics its Information Technology Laboratory has tested face recognition since the FERET programme of the 1990s and the first Face Recognition Vendor Test in 2000. Testing is voluntary and free: a vendor submits a compiled algorithm, NIST runs it on datasets the vendor never sees, and the results are public under the vendor's name.

What does it test?

• FRTE 1:1 verification: is this the same person as that enrolled image? Reported as false non-match rate at a fixed false match rate.
• FRTE 1:N identification: does this probe match anyone in a gallery of up to millions? The track closest to what a face search engine does, though NIST galleries are curated portraits and mugshots, not scraped web photos.
• Demographic effects: NISTIR 8280 (December 2019) measured false positive rates across race, sex and age for 189 algorithms and found differences of up to a factor of 100 between demographic groups for some algorithms, with the best algorithms showing much smaller gaps.
• FATE: face analysis rather than recognition, including presentation attack detection, image quality, age estimation and, since 2024, synthetic image detection.

The programme was renamed from FRVT to FRTE and FATE in 2023 to separate recognition from analysis. Leaderboards update as vendors resubmit; a result is a snapshot of one algorithm version on one dataset. Consumer engines such as PimEyes, FaceCheck.ID or FaceSearch do not appear in NIST results because they do not submit algorithms, so no NIST figure applies to them.

1. 1993

   FERET programme

   First systematic face recognition evaluation, funded by DARPA, run with NIST.

2. 2000

   First Face Recognition Vendor Test

   FRVT 2000; repeated in 2002, 2006, 2010, 2013, then continuously from 2017.

3. 2019-12-19

   NISTIR 8280, demographic effects

   Accuracy differences across race, sex and age documented for 189 algorithms.

4. 2023

   FRVT becomes FRTE and FATE

   Recognition and analysis tracks split.

5. 2024

   Synthetic image detection track

   FATE begins evaluating detection of AI-generated faces.

How do you use NIST's work?

• Read the FRTE 1:N and demographic reports before accepting any accuracy claim; see accuracy, bias and testing for how to read the numbers.
• Check whether a vendor's algorithm is in the leaderboard at all; many consumer engines are not.
• Note the dataset: mugshot and visa photos are far easier than web photos, so a NIST error rate is a floor, not a prediction for a web search.