AI that makes realistic face photos from sketches?

With the development of AI, it is becoming easier to create vivid pictures of virtual people that do not exist. A Chinese research team has developed DeepFaceDrawing, a technology that converts rough hand-drawn sketches into facial photos that capture features in real time. It is quite surprising that the hand-drawn sketch turns into a realistic face in an instant.

If you look at the video related to Deep Face Drawing, there are two canvases on the left and right. You can see a human figure drawn with several lines on the left canvas. Adding an outline here changes the picture of a woman in the canvas. If you draw thick lips, each time you draw a line on the left canvas, the body of the photo will gradually become clearer. When the approximate figure of the person for the illustration is determined, adjust the eye, nose and mouth elements by moving the slide on the upper right of the screen. Of course, not only women but also men are possible.

The original illustrations were drawn in various styles, but all of the completed photos reproduce the characteristics of the original illustrations. Deep Face Drawing was developed by a research team at the Chinese Academy of Sciences and Hong Kong Shengxi University. The technology that converts a picture drawn by deep learning into a picture regards the inputted illustration as fixed. However, this new approach learns a plausible face placement from a sketch drawn by learning, and finds an approximate location in the input sketch. In this case, sketches drawn with input information are much more flexible than fixed types. This method can create realistic faces even in inaccurate or rough sketches.

The system consists of a built-in module, characterization mapping, and photo synthesis. The built-in module has an auto-encoding architecture and five features to learn. In addition to both eyes, it is divided into nose and mouth, and sketch notifications, so you can adjust them with 5 sliders.

According to the research team, the evaluation showed high visual effect. In fact, it is explained that as a result of using the system other than the research team, desirable results were obtained with usefulness and expressiveness. Related information can be found here .