FIELD: data processing.
SUBSTANCE: invention relates to the field of data processing technologies, in particular to methods for training facial recognition by key points. The expected result of the claimed technical solution is achieved by the fact that the claimed solution provides for the possibility of obtaining a training image of a face; three-dimensional reconstruction of a training image of a face; generation of a training coordinate UV-grid; entering a training face image into a semantic segmentation network to extract a predictive coordinate UV-grid; calculating the loss intensity of a predictive coordinate UV-grid; calculating the gradient using a loss intensity indicator; defining a semantic segmentation network as a face recognition model.
EFFECT: increase in the accuracy of determining the coordinates of key points during face recognition.
13 cl, 10 dwg
