Rank One Computing (ROC) was founded in June 2015. The first NIST face recognition benchmark available for participation was the third and final phase of the NIST Face In Video Evaluation (FIVE) in December 2015. In March 2017, Grother et al. published the results of this third phase of testing , which demonstrated the benefit of ROC SDK to early adopters of the algorithms.
Some of the highlights of Rank One’s performance include:
This performance is based on version 1.3 of the ROC SDK. The submission used generic parameters for face detection, tracking and video processing. Had Rank One been able to participate in the earlier phases of FIVE these parameters would have been more precisely tuned for their environment and perhaps multiple algorithms would have been submitted, as most vendors submitted three algorithms to the final phase.
Version 1.3 of the ROC SDK was a highly compelling offering; top tier face recognition accuracy at a template size 1/20th of any viable competitor, all while running in native libraries with simple software interfaces. However, ROC SDK is currently at version 1.11, and over this time the SDK has had significant improvements:
Rank One is eagerly participating in all ongoing and future NIST Face Recognition benchmarks. These ongoing benchmarks will continue to demonstrate the unparalleled accuracy and efficiency of the ROC SDK.
 P. Grother, G. Quinn and M. Ngan, “Face In Video Evaluation (FIVE) Face Recognition of Non-Cooperative Subjects”, National Institute of Standards and Technology Interagency Report 8173}, March 2017.
The latest version of the ROC SDK, version 1.10, is now available and is complete with numerous improvements. First, the frontal face detection algorithm is more accurate, faster, and better aligned with subsequent recognition tasks. The ROCFR unconstrained face recognition algorithm is more robust on confounding races/ethnicities. Lastly, the facial pose, gender, race/ethnicity, and age estimation algorithms are now more accurate, continuing the ROC SDK's legacy as the most effective face reconition solution for multiple use-cases.
ROC SDK version 1.9 provides a significant boost in accuracy to both the ROCFR and ROCID algorithms with roughly a 5x decrease in error rates for both algorithms. Two new classifiers have been introduced: an automated race/ethnicity classifier and an automated glasses detector. This release continues the Rank One tradition of building the best face recognition solution for your needs. Already among the most efficient, accurate, and affordable face recognition solutions, the ROC SDK continues to push the limits of face recognition technology.
Rank One is excited to announce the release of ROC SDK version 1.8. The most notable improvements include an over 25% improvement in enrollment speed (i.e., the speed to generate roc-templates), and the inclusion of a new face quality metric. The improvements to enrollment speed continue to vet the ROC SDK as the fastest face recognition solution on the market and support the ingestion of large face databases or template generation on small embedded devices. The addition of a quality metric, designed to predict whether or not an image will generate a successful match, allows ROC SDK integrators to provide more meaningful feedback to system users. Version 1.8 contains many stability improvements as a result of working hand-in-hand with our integrators, making this version of the ROC SDK the most easily deployable to date.
At the Global Indentity Summit (GIS), Rank One, alongside their partners at Ideal Innovations, Inc., demonstrated the game changing efficiency and accuracy of the ROC SDK in face recognition applications embedded on laptops and mobile devices. While other technology providers require multiple servers to host large face databases, and their customers still must wait long periods of time for the search results to become available, government and industry attendees at the GIS witnessed highly accurate, large-scale searches across millions of images on laptop and mobile hardware, with near-instantaneous search results. As Rank One's user base continues to grow, the face recognition community is realizing through events like GIS that accurate and scalable face recognition solution can be licensed for a fraction of the cost of legacy solutions. Contact us today for your organization to get started with your evaluation.
Rank One is pleased to annouce the release of a new version of the ROC SDK. Version 1.7 brings our customers and integrators several enhancements, including automated liveness/spoof detection, improved face detection accuracy, and wrappers for C#, Java and Python. Our engineers are busy developing many new features and algorithm improvements, continuing the rapid pace of improvements for the ROC SDK.
Version 1.6 of the ROC SDK comes with significant accuracy improvements with respect to recognition, along with numerous optimizations for various API calls. Additionally, the ROC SDK now includes a highly accurate age estimation algorithm. We continue to polish the Explore application; among other additions, this version introduces support for dragging-and-dropping preenrolled galleries, allows for multiple face detections within our Watch functionality, and allows users to specify where galleries are stored.
Rank One's partner, Ideal Innovations, Inc. (I3), has released the Facial Automated Biometric Identification System (FABIS)-Mobile product. Powered by the ROC SDK, this solution enables searching over 1M face records on a mobile device without a backend connection and tens of millions of face records on a laptop. Users of this system will have access to Rank One's breakthrough face recognition algorithms in a solution developed and delivered by an organization (I3) with extensive experience deploying operational systems to defense customers and private industry.
Our latest version of the ROC SDK, version 1.5, incorporates major improvements in facial landmarking resulting in improved recognition performance. Face detection has been improved as well: in addition to being faster, face detection confidences are now normalized for easier interpretation. We also have introduced a new face detector designed to detect faces that are partially occluded. Our demonstration application, Explore, has been updated with many new features and has improved support on Windows. We believe the ROC SDK continues to be the premiere face recognition solution on the market.
Law enforcement and industry partcipants were high with praise after observing ROC-integrated technology performing challenging face recognition searches across large databases. Whether it was searching 50M face records on a single laptop, or 1M face records on a mobile device without a backend connection, the postive hits returned in near real-time demonstrated the unprecedented capability offered in the rapidly evolving ROC SDK.
See this capability for yourself at the Police Security Expo (June 28-29, 2016 in Atlantic City, NJ). We will be located with our partners from Ideal Innovations, Inc. (I3) at booth #175.
We are pleased to announce the release of ROC SDK v1.4, which advances a breakthrough set of commercial-off-the-shelf face recognition algorithms. This newest SDK version introduces efficiency improvements, improved face detection, new API features, and a gender estimation algorithm. A major milestone has been reached as this version supports enrollment and matching on Android mobile devices. With the efficiency of ROC SDK, a commodity phone can now perform on-device searching of up to one million face records in a matter of seconds.