China designs an algorithm that recreates human faces
The tool incorporates a deep learning neural network capable of mapping elements detected in a drawing
The Chinese Academy of Sciences (CAS) has designed a new Artificial Intelligence (AI) algorithm capable of realistically recreating human faces from basic hand-drawn drawings, even if they are incomplete.
“DeepFaceDrawing” allows portraits of human faces to be generated, and uses recent deep learning techniques for transitioning from image to image. The process is based on a sketch in pencil and without handmade color, similar to cartoons, made by people with little knowledge of drawing .
One of the study’s authors, from the CAS and the University of Hong Kong (China), explained that most AI solutions require professional sketches to generate realistic images. But DeepFaceDrawing solves this problem “by implicitly modeling the form of images from Plausible faces and synthesizing a face image in this space to approximate a drawing.”
Likewise, the technology incorporates a deep learning neural network that is capable of mapping the elements it detects in designs and generating intermediate results, which improve the information flow of the system. Thus producing better final images even if the sketches are approximate or are partially incomplete.
DeepFaceDrawing is capable of working in real time, reacting to handmade changes to the sketch and modifying the photorealistic portrait, and can be used by non-artists.
Taken and with information from El Sumario
