We are excited to announce the release of version 1.15 of the ROC SDK, the world’s most versatile face recognition solution. We self-impose an expectation to release new versions of the ROC SDK on an ongoing three-month cycle and to provide our users with accuracy, efficiency and usability improvements in each release. While our users have come to expect meaningful advances from each new version, the accuracy improvements embodied in version 1.15 are unprecedented in the life of the company, yielding an exponential decrease in face recognition error rates. This advance is on top of what was already one of the most accurate face recognition algorithms in the world.
Based on our internal face recognition accuracy benchmarking on front-facing, standards-compliant face imagery, the False Acceptance Rate (FAR) decreased from an impressive one in one million (10-6) facial comparisons, to now only occurring in one in ten million (10-7) comparisons, all while maintaining a True Acceptance Rate (TAR) of 97%. This represents a full order of magnitude reduction in error. The accuracy on unconstrained “in the wild” imagery, with natural variations in facial pose, occlusion, and illumination, increased from 50% TAR at a FAR of 10-6, to 75% TAR at the same FAR (one in one million), which represents a 50% increase in accuracy at very tight error tolerances.
While these major accuracy improvements will be noticeable by any user upgrading their platform, we are unable to report official updated NIST FRVT Ongoing results as NIST has temporarily halted submissions between February and May and did not permit Rank One to submit prior to the hiatus. We are looking forward to demonstrating our latest developments in version 1.15 along with our R&D team’s continued enhancements underway in version 1.16. Rank One will continue submitting to all NIST benchmarks, to include FRVT 1:N 2018 and FRVT Ongoing. Through these benchmarks we will continue to demonstrate the ROC SDK’s measurable versatility and performance across face recognition applications ranging from mobile to enterprise.
The significant accuracy improvements in version 1.15 will be accompanied with a modest decrease in enrollment speed, and increase in template size. However, given the ROC SDK’s 5x to 10x advantage in enrollment and comparison speeds over other published vendors, our users can remain confident they are using the most versatile face recognition solution on the market.