Scientists from South Ural State University (SUSU) trained a neural network to recognize the faces of people who are hiding from cameras, using awkward angles.
The work was successfully completed with the support of the Russian Science Foundation. We trained the VGGface neural network to recognize so-called 2.5-dimensional images. And people who are trying to take advantage of awkward angles can no longer escape from such artificial intelligence.
Unlike the standard two-dimensional system, scientists used a 2.5-dimensional one, which also adds the depth of the image (that is, the distance from the camera to the person's face is determined). To measure the depth, scientists used an estimate of the curvature of the infrared grid - infrared illumination.
If a person's face is far from the camera or too close to it, if it is turned at an angle or the lighting is "not right", then the neural network may give a false result. We solved this problem.
Earlier, SUSU student Anton Averyanov, together with his team of developers, created SheetsGPT — a neural network capable of simplifying work with Excel.
