In news programs to hide the identity of some people apply the simplest technology – pixelization of the image and distortion of the voice. This allows you to convey the essence of what the subject says, but not his personal emotions about the subject of the conversation. And they are sometimes equally important, so at the University of British Columbia they taught artificial intelligence to redraw people's faces beyond recognition, but at the same time to preserve their emotions.
The work begins with the fact that the operator manually controls the original image, clearing explicit markers, which could point to a person's personality. He also arranges tags describing the emotions of the subject. Then the AI makes a quick and simple randomization of the pixel composition so that it is impossible to accurately restore the original image if you resort to reverse editing.
Then begins the present work, where the AI copies the technique of Van Gogh, Picasso or other famous painters to draw new facial features. He is looking for a form that can combine the inner state of a person and his external manifestation. Accuracy and realism are deliberately ignored, the proportions, shades and position of individual objects begin to take on grotesque features. You no longer recognize in the picture of the one with whom it was painted, but you will understand that a person smiles sadly or is ready to yell with fear.
Frame by frame, AI can redraw any video, creating a fundamentally new content. Not only for the needs of the media, but also, probably, initiating a new direction of fine arts. Imagine a film of inmost stories from people who would never have revealed them if they were not sure that nobody would recognize them personally, but the audience will understand the whole amount of emotions and the emotional background of those events. And the technology is suitable for use in virtual reality systems.
js.src = “&version=v2.8”; 'script', 'facebook-jssdk'));