Russian start-up outsmarts Google in face recognition
10 Dec '15
The University of Washington has announced the results of an international face recognition algorithm competition called Megaface it hosted recently. N-tech.lab, a young Russian start-up, joined the global Top-5 in all the four nominations of the contest, and won two of these, outshining Google in the process, the Russian business daily Vedomosti reported.
Under terms of the competition, each participant was expected to identify photos of people from two collections in a large database. Collection one contained about 100,000 pictures of people of the same age, while in collection two there were as many as 900,000 pictures of people of a range of ages. All in all, a database of one million images was to be looked into.
The host offered 100 participants four nominations. Each team could participate in one of two categories. In category one, teams were allowed to use as many photos as they wanted to ‘teach’ their algorithms; in category two, only up to 500,000 photos could be used for that.
An algorithm developed by the Russian company is reported to have been able to recognize more than 73% of faces from collection one in the million face database, outclassing Google with its just over 70%. In category two, in which a 500,000 picture limit was imposed for teams to ‘teach’ their algorithms, N-tech.lab was the runner-up. Google also outperformed competitors in recognizing 900 images from collection two without any picture limit with 74%. The Russian start-up ended up second in this part of the contest as well with 52%. However, in the search of 900 images in category two (with a 500,000 picture limit imposed for teams to ‘teach’ their algorithms) the Russian team beat all competitors.
“We consider it a victory, as our algorithm performed best in the main category. We’re pleased to observe that we used only three computers to ‘teach’ our algorithm, while Google needed a thousand servers to do the job,” N-tech.lab founder Artyom Kukharenko said.
N-tech.lab was set up earlier this year; the founder is an alumnus of the Moscow Lomonosov State University (MSU). The start-up has short-term plans to release a number of commercial products that will use its winning algorithm. One of those will serve the entertainment market segment as part of a dating service, enabling users to search for people with given looks.