Neural network learns to create three-dimensional models of human faces using single photo
A group of researchers from the University of Southern California has created a program that can create the realistic three-dimensional image of a human face using a single photo.
According to N + 1, high-quality photos that are taken at several angles under ideal lighting are usually used for creating “virtual masks”.
This is necessary in order to transfer all the features of a human face, including asymmetry, curves and fine details, including moles.
The method, developed by researchers from the US, makes it possible to use only one non-professional photo taken in the everyday environment for this purpose.
The researchers created a model that could “read” the shape of the face in the photo. Then they used a previously prepared the convolution neural network VGG-19 which analysed the face workpieces and picked up realistic textures for them.
To accomplish this task, the neural network used a database that contained the photographs of men and women of different ages and different nationalities.
According to the developers, in the future, the new technology can be used for creating VR-avatars and developing online games.
