Latest NIST FATE sees two new algorithms, fresh focus on interocular distance

The latest results of the National Institute for Standards and Technology (NIST)’s Face Analysis Technology Evaluation Age Estimation & Verification (FATE AEV) are up, and they include two new algorithms: one from the UK’s Yoti (yoti-004) and one from Swiss firm Privately (privately-000).

The results page also now highlights the metric of Mean Absolute Error (MAE) by Interocular Distance, referring to the distance between the centre of the eye pupils measured in pixels using mugshot images.

In a post on LinkedIn, Robin Tombs, CEO of Yoti, breaks down the metric – and how Yoti fares in the evaluation.

Mugshots at 480×600 pixels are “the NIST images that best approximate selfie face images captured on mobile phone cameras,” Tombs says. Pointing to NIST’s results, he notes that in the category of Mean Absolute Error (MAE) by Interocular Distance, testing on 917,582 mugshots for unique white 18-30 year olds captured across the U.S. shows Yoti’s algorithm yoti-004 to be the most accurate.”

“NIST reported MAE for yoti-004 white male 18-30 year olds (with IOD of 120-199 pixels) is 1.71 years.”

Tombs says the evaluation validates its own internal testing results, which opponents of facial age estimation have dismissed as untrustworthy – but which showed similar results to NIST.

Regula repeats as top-ranked algorithm across geography

The rest of the table evaluating MAE for interocular distance shows Innovatrics’s algorithms in the number two and four slots. Regula takes third, and Idemia sits at fifth.

In NIST’s ranking of MAE across geography, Regula retains the top spot, with Yoti at third and Cognitec in second. Idemia and Innovatrics round out the top five.

Evaluating MAE by dataset and variability, in the mugshot category, Innovatrics takes the top two spots, with Idemia at three, Yoti at four and ROC at five.

Yoti, Privately see success with FAE products

Yoti continues to see major deployments of its FAE product. TikTok recently enlisted Yoti to provide age estimation, joining Meta’s Instagram and Facebook among social media platforms integrating the service. Minecraft will deploy Yoti’s FAE for access to its chat feature.

The UK-based firm Yoti recorded a 62 percent year-on-year growth in revenue in 2025, reaching 29 million pounds (about 39 million dollars). The firm’s visibility surged in July around the deadline for the UK’s Online Safety Act, flirting with the top of the app charts. In November, it announced the completion of a billion age checks.

Privately, meanwhile, has deployed its device-based age assurance tools in more than 100 countries. In May 2025, the firm announced it was processing nearly 300,000 age checks monthly through its AgeAI and FaceAssure products. With headquarters in Lausanne, it is positioning itself as a local provider should Switzerland adopt online safety laws in line with other EU nations. Its algorithm placed in the top 20 for geography in NIST’s latest rankings.

Posting on LinkedIn, Privately CEO Deepak Tewari celebrates the firm’s first submission to the NIST leaderboard, declaring himself “pretty chuffed.”

“I am happy to report that privately-000, our FAE model aimed at younger demographics, clocks outstanding accuracy for under 17 and good accuracy overall – all within a model size <20MB,” he writes. “We have squeezed the performance that competitors are delivering on huge server based models into a very light-touch model.”

The model, he says, can run fully on a user’s browser without the need for any facial image to leave the device. “The NIST test further shows that we have the best in class inference times around 64 milliseconds which is conducive to running low-latency computations on the edge (on-device/on-browser) implementation.”

Article Topics

biometric testing  |  biometrics  |  Face Analysis Technology Evaluation (FATE)  |  face biometrics  |  facial age estimation (FAE)  |  NIST  |  Privately  |  Regula  |  Yoti