Having looked at the photo, any person can easily imagine what is depicted on it as a three-dimensional image. And even if some of the subjects remain behind the scenes (for example, we see only the front part of the back of the chair, but do not see the back), it is not difficult for us to guess how the object will look on the reverse side. For AI, such tricks are complex, but recently a group of engineers from the DeepMind project not only taught the artificial mind to do the above, but also to translate the 2D image into full-fledged 3D.
The new system was called the “Generative Query Network (GQN) . To teach the artificial mind to analyze two-dimensional scenes, the team of researchers showed him images of one scene from different points of view. This chain of images AI used to understand the principle of changing objects when changing the angle. Moreover, this approach allowed us to learn how to predict how the scene from a different angle of view will look, taking into account not only the position of objects, but also the lighting. As reported by one of the authors of the work, Ali Eslami,
“Imagine that you are looking at Mount Everest and moving about a meter. The mountain will not change its size, which will give you information about the distance to it and its size. And if you look at the mug and do the same – its position will change. This is how our technology works. “
“Moreover, the algorithm is trained in the same way as a person. Having seen the same object many times, the AI analyzes its characteristics, remembers and then uses it in a second interaction. Now artificial intelligence can recreate a whole labyrinth by scanning just a few photos taken from the inside. “