Rank One delivered another impressive performance in the latest iteration of the NIST FRVT Ongoing face recognition benchmark. While nearly every vendor had gaps in their algorithmic performance, Rank One’s v1.18 algorithm did not have a single performance deficiency. Read more about it on our blog.
Rank One Computing has released version 1.18 of the ROC SDK, its flagship facial recognition solution, which is delivered with substantial accuracy improvements and several new features. Now available to Rank One’s user community, this release is one of the most significant progressions ever to Rank One’s face recognition algorithms, which are already industry-leading across most performance measurements.
The main highlight of this release is up to a 2x reduction in error rates across all types of facial imagery (constrained and unconstrained). This improvement was particularly profound across the more challenging demographics (race and gender). Continue Reading...
A key new feature added with version 1.18 from Rank One is emotion classification. The company’s SDK users will be able to quickly assess whether a face appears joyful, sad, angry, fearful, disgusted, surprised, or has a neutral expression.
In previewing the new release, Rank One’s CEO Brendan Klare states: “Version 1.18 represents yet another major progression delivered to our customers and partners. Through a lot of hard work and innovation, our pace of our accuracy improvements is accelerating with no end in sight. Additionally, we are further re-investing in long term initiatives to ensure the ROC SDK remains an industry leader in accuracy, efficiency, and ease-of-use. Best of all for ROC customers under maintenance plans, our advanced licensing policies ensure that they can automatically take advantage of all these v1.18 enhancements.”
A detailed description of the improvements delivered in this release can be found in our blog post.
CEO Brendan Klare was a featured speaker last week at the NIST IFPC 2018, an international face recognition conference sponsored by the Department of Homeland Security’s Science and Technology Directorate. Dr. Klare was one of an exclusive set of speakers from across the globe involved in face recognition development, deployment, and operations; his presentation addressed emerging applications in commercial face recognition solutions which was informed by the wide range of applications and systems being developed by Rank One’s customers.
IFPC 2018 was focused on all technical factors affecting the deployment and use of high performance face recognition applications, including applications, standards, advanced and rapid capture, quality assessment, age and ageing effects, demographic effects, datasets, their preparation, training and tuning, presentation attack detection, non-cooperative uses, accuracy measurement, and performance tests. Part of the Department of Homeland Security’s Science and Technology Directorate, the conference aims to assemble a set of speakers from across the globe involved in face recognition development, procurement, deployment and operations. The overarching goal is to bring greater maturity to face recognition by improving performance, transparency, and trustworthiness.
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 ourperformance perfectly: the accuracy is within a percentage point of the top performing submissions, and the speed is 10x faster. Continue Reading...
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.
We’re launching today our comprehensive blog to share industry information and education with our community of engineers, integrators, and industry insiders, with the focus on acting as the ‘go to’ resource for the biometric industry.
Designed and developed by ROC executives, the solution-based blog helps industry professionals navigate facial recognition concepts, protocols, and other helpful details from an expert yet easy-to-understand perspective. The inaugural posts answer common questions like "How does face recognition actually work?" and "What is a False Accept Rate?" Continue Reading...
"We’re excited to launch this best practice e-learning resource for the facial recognition community,” said CEO/Co-Founder Brendan Klare. “This blog will leverage our depth of knowledge, experience, trend identification, and ongoing research and analysis to assist the face recognition and biometric community at large.” Find the new blog at blog.rankone.io.
ROC SDK version 1.17 is now available. With this new release users will receive more accuracy enhancements, faster processing speeds, and usability improvements.
The biggest improvement with version 1.17 is another order-of-magnitude reduction in false positive error rates when operating on unconstrained, or “Wild”, imagery. These significant upgrades in accuracy will be particularly noticeable to integrators that process imagery with variations in facial pose and environmental illumination.
In addition to large accuracy improvements, processing speeds for v1.17 are roughly 10% faster than v1.16’s industry leading speed and efficiency. Continue Reading...
The final major improvement delivered in v1.17 is a significant extension to the video processing API, which was enabled by the Rank One team porting a large amount of the source code from our Explore GUI system into the SDK and making these features accessible through a well thought-out API. These enhanced features will reduce the overhead required to develop tracking and monitoring applications using the ROC SDK.
When asked about the achievements delivered in ROC SDK v1.17, Chief Engineer Scott Klum said, “Rank One's core algorithms have always been on the cutting edge of accuracy and efficiency. With version 1.17, we had a number of breakthroughs that simplified algorithm architecture while, in keeping with Occam's razor, providing significant accuracy gains. These improvements give our users continued access to a technology so efficient that it can run on an embedded device, with accuracy comparable to (and in most cases superior to) competitors’ technology that must be run on a server.”
Rank One is not letting up on algorithm R&D anytime soon. Users who have become accustomed to major quarterly enhancements to the SDK can expect another big boost in accuracy that is already underway and scheduled for release in mid-December.
Rank One is pleased to announce the release of version 1.16 of the ROC SDK, which includes another round of significant feature enhancements to the world's most versatile face recognition solution.
Several accuracy improvements have been made to the SDK in this release. First, version 1.16 provides across-the-board precision improvements for all capture scenarios (constrained cooperative, unconstrained cooperative, and unconstrained uncooperative). These accuracy improvements are on top of the major improvements released in version 1.15. Second, face recognition accuracy is now more equally balanced across different racial and gender cohorts. Finally, major enhancements to the face detection and recognition framework now allow users to perform recognition across all facial pose angles, including full-pose (-90° to 90° yaw) face detection and profile to frontal face matching. Adding "full-pose" face recognition capabilities is the first step in a series of scheduled enhancements that will enable our users to achieve successful facial matches despite extreme facial pose angles and low quality capture environments. Continue Reading...
Perhaps the most significant improvements in this release relate to algorithmic efficiency. Previous versions of the SDK were already proven to be the world's most efficient face recognition solution in terms of comparison speed and enrollment speed, per testing from the National Institute of Standards and Technology (NIST). In version 1.16, template size has been reduced by 60%, down to 133 bytes. The ROC SDK's small template size, unparalleled comparison and enrollment speeds and robust accuracy are the result of the Rank One team's relentless research, and in turn yield significant hardware cost savings for customers and enable on-edge applications that would otherwise be impossible.
Rank One is also excited to release major enhancements to its passive liveness/anti-spoof algorithm. Rank One's patent-pending technique requires no special hardware and is effective at deterring printed photo and digital screen replay attacks using a single image capture, meaning powerful liveness/anti-spoofing can be performed passively, without requiring multiple images, multiple poses, user response or specialized hardware. This liveness solution performs well on both fixed-focus and auto-focus cameras, thereby covering both front-facing and rear-facing cameras on mobile devices and addressing both selfie and access control use cases.
As always, there are several enhancements to the usability of the SDK. Ease-of-integration is one of the key intangible reasons the size of Rank One's customer base continues to accelerate. Rank One is committed to continued excellence on the ease-of-integration front, whether it is through our stable well-thought-out API, support for multiple programming languages, command line applications, or responsive customer support provided by the same engineers who develop the technology.
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. Continue Reading...
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.
Rank One Computing recently released version 1.14 of the world's most versatile face recognition solution, the ROC SDK™.
Rank One's CEO, Brendan Klare, had the following to say about this latest release: "We are excited to deliver to our customers yet another significant accuracy improvement. We hope this latest version, coupled with the measured performance in the NIST benchmarks, will attract new users who are seeking a face recognition solution that provides both top-tier accuracy and efficiency. Furthermore, while accuracy and efficiency can be quantitatively measured through third-party testing, what cannot be measured is usability and customer support. We have a burgeoning reputation for the support we provide to our customers, as well as the usability of our SDK. Every aspect of this experience, whether it is accuracy, efficiency or usability, will continue to improve over time." Continue Reading...
The new algorithm provides substantial accuracy and efficiency improvements over previous versions across all types of imagery and environments. These major accuracy and efficiency improvements will be reflected in the coming days as part of the NIST FRVT Ongoing benchmarks, where Rank One's v1.12 algorithm is currently the leading submission in combined accuracy and efficiency.
Rank One was represented at the KNOW conference through participation in a discussion panel, as well as numerous Rank One customers sponsoring the conference.
Rank One hosted a booth at the 2018 FedID conference where CEO Brendan Klare, COO David Ray, and Chief Engineer Scott Klum engaged with current and prospective customers as well as other partners and integrators regarding the key strengths and differentiators of the ROC SDK. Common feedback provided from Rank One’s existing government and law enforcement customers was that they benefited tremendously from the continuous accuracy improvements coupled with easy access to these quarterly improvements, as well as the world-class speed of Rank One’s software.