AI learns how to take a simple sketch and turn it into a completed portrait

China has created an advanced analogue of pix2pixHD and Lines2FacePhoto programmes. The mission of the neural network is to help artists with their work.

Scientists at the Chinese Academy of Sciences and the University of Hong Kong have created a DeepFaceDrawing programme, a technology designed for the artistic process of drawing. By taking a simple sketch, a neural network can turn it into an extraordinary photorealistic portrait with numerous details.

The principle of the programme is as follows: the image is divided into basic elements and then carefully analysed. Following that, the neural network chooses similar parts of real faces from its database and composes them. The outcome is a more balanced portrait that looks like a modified version of the original drawing.

The quality of input data for development doesn't matter: DeepFaceDrawing can work with even the sketchiest of images, modelling realistic faces from them, that open vast new possibilities for the neural network.  The video below displays the programme's features.