Scientists from the United States of America have developed an algorithm based on neural networks that can “liven up” people captured in photographs. The algorithm creates a 3D model for the human figure and then reproduces the specified animation with it.

When we look at the images with people caught by the photographer at the time of the movement, we usually have an idea about how these people moved a moment after the camera was released. For software algorithms, this task is incredibly difficult. But it is quite realizable - proved the staff of the University of Washington, as well as representatives of the company Facebook participating in the project.

How the results of the unique neural network look like can be seen in the video below.

“In fact, a bunch of several complex algorithms is used here, because there are also several tasks in the process - you need to recognize and separate a person from the background, build his three-dimensional model, cover it with texture and animate,” says the project author, Ira Kemelmaher -Slitserman.

It is noted that the algorithm exists in several versions - for example, for augmented reality, in which the results of its work look particularly impressive.

planet-today.ru