Rank One claims a victory in NIST’s new benchmark for biometric face-scan memory use

A key U.S. government standards agency has begun tabulating runtime memory use by facial recognition algorithms, and a Colorado-based software maker is claiming bragging rights on its rank.

The National Institute of Standards and Technology publishes an ongoing report on benchmarks for any facial recognition algorithms submitted in an effort to help define the rapidly evolving market.

In January, NIST added the peak-memory benchmark to its Face Recognition Vendor Test because the environments into which the technology is expected to grow quickly — mobile, embedded and on-edge applications — generally have little random access memory to spare.

Executives at Rank One Computing, a maker of facial recognition algorithms, wasted little time in promoting the fact that their algorithms ranked third and fifth in a field of 190 competitors.

The new NIST benchmark measures “the peak size of the resident set size logged during enrollment of single images.” This makes it valuable for developing mobile, edge, and embedded applications, which typically face significant resource constraints, such as low memory.

Two Rank One submissions, named rankone 007 and rankone 008, with 67MB and 79MB, respectively, of RAM (page 22, lines 121 and 122 in the January report).

Algorithms requiring less memory were from TUPU Technology Co. Ltd, which was first with 33MB; Videonetics Technology Pvt. Ltd., second with 61 MB; and Ayonix, fourth with 69MB. According to the company, the median use recorded in the report was 730MB.

Rank One executives say the other three offerings are many times less accurate than theirs. The company also says its enrollment speeds are more than twice as fast as the median algorithm speed, and it makes another algorithm which requires less than 10MB of RAM.

The January publication (NIST’s sixteenth such report since 2017) is still a draft and is open for comment.

Article Topics

algorithms  |  biometric testing  |  biometrics  |  biometrics at the edge  |  Face Recognition Vendor Test (FRVT)  |  facial recognition  |  mobile  |  NIST  |  ROC  |  standards