AI Google learned how to create a 3D scene from images

In 2014, the British startup DeepMind was bought by Google. And this is the case when the purchase can be called successful. Today DeepMind demonstrates breakthrough technology in the field of artificial intelligence, and one of these is Generative framework Query Network (GQN), which now will be discussed.

Typically, neural networks based on images, using a database with lots and lots of pictures. To artificial intelligence in the future could identify objects in the pictures, need the pictures from the database manually mark objects, for example, sofa, floor, chair and so on. This process is quite tedious and time consuming.

So Google invented a new, more sophisticated approach. Its essence lies in the fact that artificial intelligence will identify the objects to augment the scene, to understand what actions occur in the frame, to take into account the lighting, the materials of various objects. Now for creating 3D scenes from 2D images will only need to show artificial intelligence image of the scene, and you’re done.

Below you can see how the AI based on one image builds a scene:

The project is in its early stage of development, and talk about complex scenes is not necessary. But even now the result is stunning.

Share your opinion in the Telegram chat or in the comments below.

According to the materials of deepmind

Leave a Reply

Your email address will not be published. Required fields are marked *