Transfer learning based on the EfficientNetV2B0 neural network architecture in the problem of constructing a 3d human facemask

The paper uses the transfer learning approach to retrain the EfficientNetV2B0 artificial neural network architecture in the problem of creating a 3D human mask. The developed model predicts 478 points of a person's face, each of the points is represented as a vector of three coordinates. A method of data labeling and an approach to learning are presented.