Researchers from the Chinese Academy of Sciences and the City University of Hong Kong collaborated on and introduced a new AI tool that can generate realistic human portraits from simple sketches.
Unlike previous deep learning algorithms, the new software called DeepFaceDrawing treats input sketches as ‘soft’ constraints that will guide image synthesis. As a result, it can produce high-quality face images from a rough or incomplete drawing.
Image-to-image translation techniques are not new to the tech industry. A generation of researchers has continuously churned out AI software that can achieve the same result. However, the difference is that previous algorithms can only produce images if the input has high-quality artistry or is accompanied by edge maps.
In their user study, the researchers confirmed the tool’s usability and expressiveness. They describe DeepFaceDrawing as easy-to-use, even for non-artists. The source code is still in the works and is yet to be released.