9 Sep '16
Vocord, a Russian developer of advanced , beat all competitors, including a Google algorithm, to become the world’s best during tests on the MegaFace online testing grounds, RVC, Russia’s fund of funds for innovation.
The MegaFace online testing grounds are a database of a million photos of more than 690,000 different people. The pictures were taken in natural conditions. People were not posing, their facial expressions were different, and so was illumination. The competition was aimed at evaluating the accuracy of recognizing faces from a target group amidst an ever-growing number of “distracting faces” (anything between ten and a million such faces) from the MegaFace database. The MegaFace test is said to enable the assessment of existing face recognition algorithms using a huge number of different images.
VOCORD’s proprietary algorithm, based on deep neural networks, is reported to have shown the best results globally in ranking 1 face identification category. That means that software is expected to be precise in selecting the right face image out of a million test pics, ranking it number 1 by its match to a reference database of people. As a result, the VOCORD algorithm was correct in recognizing 75.127% of faces, the world’s best achievement as of today, RVC emphasized.
The Russian recognition algorithm is said to be able to conduct up to three million comparisons per second.
Vocord is a portfolio company of S-Group Ventures and Lider-Innovations, two Russian funds RVC once invested in to set up.