After learning, we are talking about an algorithm that can create a realistic head model using the selfie cameras of an ordinary phone.
The algorithm itself consists of two main parts: a hypernet that learns a universal model of head and facial expressions, as well as a personalized model – it is created on a universal basis and trains the model for a specific person.
It is reported that the data set was compiled with the participation of 255 people of different genders, races and ages. In total, as part of the shooting, the experts received 3.1 million frames – they were used for training.
Note that the creation of personalized models is done using the iPhone 12. This model was chosen because its front camera produces not only the RGB layer, but also the depth layer.