Rank One Computing achieved yet another round of sweeping improvements in the most recent NIST FRVT Ongoing Benchmark, reducing error rates by between 10% and 50% on all datasets while improving algorithm speed at the same time.
Figure 2 in the report, a scatter plot of accuracy vs. algorithm speeds, summarizes our performance perfectly: the accuracy is within a percentage point of the top performing submissions, and the speed is 10x faster.
Rank One’s U.S. owned and developed algorithms continue to improve at the same pace as the rest of industry, all while continuing to deliver the fastest and most efficient implementations that save our customers time and money. In fact, in the FRVT Ongoing November 2018 report the current “rankone-005”` algorithm, which corresponds to version 1.17 of the ROC SDK, would have been the single most accurate algorithm on the mugshot data just 7 months prior in the April 2018 FRVT Ongoing report, while being 10x faster than the top algorithm in the April report (785ms vs. 71ms enrollment speeds).
This combination of top-tier accuracy and order-of-magnitude superiority in efficiency is critical to developers of systems that process streaming imagery or run on mobile devices. Additionally, our law enforcement and government customers are able to quickly re-enroll their large databases, many consisting of tens of millions of images, and quickly tap into these impr ovements. Rank One customers realize the advantage of our unique perpetual licensing model, which provides continued access to all algorithm releases with a nominal maintenance fee.
In every release to date, Rank One has led the industry in redefining speed and efficiency standards while maintaining top-tier accuracy. More details to come, but for users who prioritize accuracy above all else, we have something new and exciting coming with our next release that is scheduled for mid-January.