A new deep learning system has been developed that can generate realistic faces in real-time from simple sketches. The non-profit research company OpenAI recently introduced an artificial intelligence (AI) that can independently generate new music. The goal of the research project, which was financed by Microsoft and Tesla founder Elon Musk, among others, is not the production of music itself, but the acquisition of new knowledge in the field of AI and its benefits for humanity. Scientists from the University of the Chinese Academy of Sciences (UCAS) have now presented a deep learning system called Deepfacedrawing in the arXiv journal that creates realistic faces from simple hand sketches. The software uses a database of real human faces to generate a new face based on the sketch.
The software does not require details such as skin, eye, and lip color, as well as eyelids and other features. The demonstration of the editor shows that the software compares the strokes of the sketch with its database in real-time and generates a new face based on the line drawing. The result can still change significantly during the creation process if additional details such as hair are added by the human or if the facial shape of the sketch is adjusted afterwards. Due to the unique feature of real-time graphic creation, the resolution of the images is currently limited to 512 × 512 pixels, which restricts the practical usability of the generated faces.
The results also show that the database of real faces mainly consists of South American and Caucasian-looking women and men. It is unclear whether the Chinese scientists deliberately chose this limitation because generating faces of different ethnicities and age groups based on simple sketches was still too complex, or whether the necessary training data has not yet been integrated. This new technology has the potential to revolutionize the way we create and design faces in various industries, including gaming, animation, and film.
