Sogou biometric facial recognition algorithm hits high score in MegaFace Challenge
The MegaFace Challenge, billed as the world’s first large-scale facial recognition algorithm competition, has announced its latest results with the FaceScrub data set.
A facial recognition algorithm from Sogou scored the highest true acceptance rate (TAR) among more than 80 entries, at 99.939 percent. In addition to the high-scoring algorithm identified as “AIGROUP – Sface,” a second Sogou algorithm placed sixth with a 99.2 percent TAR. The challenge, hosted by the University of Washington, deploys algorithms to the MegaFace million-scale dataset and tests them against one million distractors, according to Sogou’s announcement.
“Face recognition is a highly competitive field in AI,” notes Sogou AI Research Director Xuefeng Su. “Our stellar performance in the MegaFace Challenge further demonstrates our leading R&D capabilities in computing algorithms and our ability to effectively allocate resources. To further advance our AI strategy, which centers on natural human-machine interaction and knowledge computing, we will accelerate and diversify the commercial applications of our face recognition technology, such as to our image search function as well as to other services.”
“In many areas, machines and humans are close to reaching performance parity,” says Sogou CEO Xiaochuan Wang. “In the field of face recognition, machines are even starting to supersede human capabilities. I’m incredibly proud of the accuracy and efficiency that our talented team of AI researchers has achieved with Sogou’s face recognition algorithm. Together with our existing AI strengths in computer vision, voice recognition and machine translation, we will further expand our AI capabilities and continue to push the boundaries of AI.”
CyberLink Corp’s FaceMe Facial Recognition Engine placed in the top 20 with a TAR of 98.41 percent, and a false acceptance rate (FAR) of less than 10-6 (0.000001).
“Today’s MegaFace Challenge’s announcement from the University of Washington affirms CyberLink’s commitment to its high-end research capability in facial recognition technology on a worldwide scale,” said Dr. Jau Huang, CEO of CyberLink. “CyberLink strives to develop the most accurate and flexible cross-platform facial recognition algorithm with FaceMeTM, to strengthen smart life experiences with domestic and international system integrators.”
FaceMe’s AI-based facial recognition can be integrated into Windows, Linux, Android, iOS, and other platforms, the company says, and run smoothly on systems with lower computing power, enabling price-sensitive mass market deployments.
Algorithms from Samsung’s research institute in Beijing and SenseTime placed second and third in the challenge, respectively.