Artificial intelligence can now create a 3D model of a person with just one short video. The technology was developed by researchers from the Institute of Informatics of the Max Planck Society and the Braunschweig Technical University.

To build a model, you need a video where the person rotates around its axis by 360 degrees with the hands apart. First of all, the neural network separates the silhouette from the background (so the background should ideally be contrast and homogeneous), then it determines the position of the joints and builds an approximate model, and finally, details it, imposing the textures of hair and clothing.

The developers plan to present the development in June 2018 at the CVPR conference. So far, as they note, artificial intelligence creates models with an average accuracy of 5 mm; The greatest difficulties for him cause skirts and long hair. But at the same time, ready-made models can be edited (made slimmer, dressed in different clothes) and made to move: run, walk and even perform pirouettes.

robo-hunter.com