The world’s best face recognition algorithm is made by Skolkovo resident Vocord, according to the MegaFace platform, which rated it above Google in one of its rankings this week.
Vocord's algorithm was developed to identify faces in crowds and public places. Photo: Vocord.
In the ranking’s first category, identification accuracy with one million distractors, the Vocord-deepVo1.2 technology was ranked number one, with accuracy ratings of more than 80 percent. Its closest competitor, GRCC, had scores of about 77 percent. Vocord also took fourth place in that category with its Deep Vo1 algorithm, which had scores of over 75 percent. Google’s FaceNet v8 was ranked in tenth place.
Vocord’s algorithm is based on deep neural networks, and also topped MegaFace ratings earlier this year. 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.
In the verification category, Vocord was ranked eighth – four places behind Google – but has closed the gap on the U.S. tech giant since the previous ratings in September.
In addition to face recognition technology, Vocord, a resident of Skolkovo’s IT cluster, makes systems for recording traffic violations and monitoring traffic volumes and for use in video surveillance. Its software is used in more than 70 projects aimed at improving safety in cities, both in Russia and abroad.
After Vocord’s success in the earlier MegaFace rankings, Alexei Kadeishvili, the company’s CTO, told Sk.ru that his team was continuing to work on optimizing their face verification algorithms, which can be used in security procedures, among other things.
“From the very beginning, we optimized our facial recognition algorithm for use in identifying a person in a crowd,” he said at that time.
“We have some very interesting results related to counteracting fraud using facial authorization,” he added.