If you fear a future in which machines will dominate humans and put them into three-dimensional simulations of reality like a video game, as in the movie “The Matrix,” this news may not be very comforting. An artificial intelligence has been able to give 3D shape – that is, depth and texture – to low resolution 2D shots.
The system was created by researcher Qifeng Chen, from Stanford University, USA, in partnership with Intel. He and his team built a neural network capable of synthesizing street photos. First, this I.A. Was fed with 5,000 low-resolution public highway images from Germany to “learn” their standards.
Then the scientists created a 3D model for each image. The task of I.A. Was to couple the two-dimensional image to the three-dimensional model. In other words, the challenge was to make the machine understand the shape, depth, distance, and texture of each object in an image.
It is as if the machine had the model of a human face in one hand and a photo of that face in the other: the goal was to make her “dress” the face with the photo as if it were a mask, understanding where the nose, , The texture of each one, the volume and depth of each.
In that case, however, the idea was to wear a street model with a photo, identifying the position of traffic lights, cars, trees, pedestrians, and everything else. The result was that I.A. Not only was he able to reproduce all these 3D photos, but he was able to create scenes, sequences of images almost as realistic as the work of a human designer.
Here’s a sample video of what the machine was capable of:
The goal of scientists, however, is not to create the Matrix. The system is still not perfect and has difficulty building photorealistic scenes. But this technique can and should be used by game developers in building 3D scenarios for games, special effects in movies, animations or virtual reality, as explained by Engadget.
The benefit of using a neural network for this is that a developer will no longer need to put dozens or hundreds of designers to spend months building movie or game scenarios, but may leave it to a smart computer to do at least a part Of work.
According to Chen, this technology is not yet ready to completely replace teams of graphic artists and high-performance machines used by studios worldwide, but it can help at least sketch how the final product will stay.