Skolkovo startup gets investment to test cure for HIV1586
French sports retailer Decathlon looks to startups to retain its pole position1051
Skolkovo joins forces with Russian Space Systems987
Skolkovo startups set sights on the skies at MAKS air show786
Dauria satellites blast off into space from Baikonur770
MAKS air show 2017604
Scooter sharing scheme expands from Skolkovo to Finland475
Cancer diagnosis startup raises funds via Skolkovo investment service393
Dauria satellites blast off into space from Baikonur0
Skolkovo startups set sights on the skies at MAKS air show0
Scooter sharing scheme expands from Skolkovo to Finland0
Skolkovo startup gets investment to test cure for HIV0
French sports retailer Decathlon looks to startups to retain its pole position0
Skolkovo joins forces with Russian Space Systems0
Cancer diagnosis startup raises funds via Skolkovo investment service0
MAKS air show 20170
Vocord, a resident company of the Skolkovo Foundation’s IT cluster, has been ranked the world’s best face recognition engine for the third time in the MegaFace Facescrub 2016-2017 challenge.
The company’s Vocord-deepVo3 algorithm topped the ranking in two of three categories, and took second place in the third category of the rating, compiled by researchers at the University of Washington.
Vocord's tech has consistently topped the MegaFace challenge, beating rivals including Google. Photo: Vocord.
In both the first category – identification accuracy with one million distractors, in which the systems are tested on databases of more than a million photos – and in the identification rate vs. rank category, Vocord’s deepVo3 took first place with accuracy rates that were consistently over 91 percent – nearly 10 percent higher than its closest rival, YouTu Lab – while deepVo1.2 took fourth place and deepVo1 took ninth place. Google’s FaceNet v8, for comparison, was ranked 17th in those categories, with accuracy rates of just over 70 percent.
In the third category, verification, deepVo3 was ranked second with an average of 94.963 percent, less than one percent behind the winner, DeepSense V2, which ranged from 94.984 percent to 95.993 percent. Google’s FaceNet v8 was ranked seventh in the verification rating, with just over 86 percent.
Vocord’s winning algorithm is already in active use on the market.
“That’s the algorithm that we use in our commercial products,” Alexei Kadeishvili, Vocord’s technical director, told Sk.ru . “It doesn’t just recognise faces better; we have significantly increased its speed, which isn’t measured in the MegaFace rating. Now we can evaluate 1,600 faces per second on one standard server, and we believe we can improve further on this,” he said.
Vocord, whose algorithms are based on deep neural networks, topped the MegaFace rating twice last year, and is constantly refining its system. MegaFace tests face recognition technology on an online database of more than 1 million photos depicting over 690,000 people photographed without posing and in natural lighting conditions.
Igor Bogachev, head of Skolkovo’s IT cluster, said Vocord’s 3D identification project has received significant support from Skolkovo, including a grant of 20 million rubles ($350,000).
“The system was created with the use of the team’s unique developments in the field of biometric identification, and search and recognition methods in video images,” he said.
“The project is primarily designed for security purposes, including at strategic sites, such as the metro, airports and other places. We are already seeing a lot of demand for such solutions, both in Russia and abroad. Together with TAL Secure Systems, for example, Vocord has begun promoting its intellectual solutions on the international market: India, and – in the near future – Middle Eastern and Gulf countries,” said Bogachev.
“I believe that the Vocord team’s experience is an excellent example of a hi-tech product created in Russia that outstrips its foreign analogues and enables us to boost security in our cities,” he added.
In addition to face recognition technology, Vocord makes systems for recording traffic violations and monitoring traffic volumes. Its software is used in more than 70 projects aimed at improving safety in cities, both in Russia and abroad.