New algorithms create better avatars

Researchers at the Swiss Federal Institute of Technology in Zurich (ETH) have developed an algorithm based on artificial intelligence to much more easily create lifelike virtual humans, known as avatars. Instead of modelling every last detail, they use intelligent algorithms that learn to automatically render animated avatars in every conceivable pose by observing 3D images of humans in just a few poses. This work is being published at the International Conference on Computer Vision.

Xu Chen, a doctoral student at ETH and lead author of the study, has reportedly developed a computer model that is for the first time able to easily represent new motion patterns. However, the new full-​body avatars cannot yet be personalized, but Chen hopes to develop the computer model so that it can create new identities at will. Like Chen, Marcel Bühler uses intelligent algorithms to create new animated faces from a combination of a 3D face model and a large collection of portrait photos.

Chen does not think there is a danger that the new technology will allow even more realistic deepfake videos to circulate, of politicians, for example. “Anyone who looks closely will still find artifacts.” Making the research on 3D rendering techniques, as well as their vulnerabilities, publicly available could help cybersecurity experts more easily detect deepfake videos on the web.

For interactive virtual reality applications, the work of these ETH researchers represents enormous progress. It is quite possible that tech companies like Facebook and Microsoft will implement the newly developed techniques in their avatars.