The neural network learned how to create a detailed virtual city

Нейросеть научилась создавать детальные виртуальные города

In order to create a truly realistic and filled with virtual world specialists in the field of programming you have to spend a lot of time, energy and effort into the process – or had, given the current success of NVidia in this field. Today NVidia a separate team involved in the development of special algorithms of artificial intelligence and neural networks, introduced its new development, which is a set of neural networks that can create their own virtual space on the basis of the viewed and analysed the video of any format, volume and type – the virtual world looks very alive!

Unlike traditional algorithms, where professionals must first program them to the rendering of content with specific instructions, the test algorithms are carried out directly artificial intelligence and neural networks. Denoting its development under the name PyTorch, experts define these behaviors as a really effective base for creating and populating a virtual space that becomes possible through the use of special graphics cards NVidia Tesla V100 technology DGX-1, that allows you to draw really lively and full of space.

As an example, the specialists of the company “fed” their system tons of information and visuals about various American cities, using data obtained from systems Apolloscape and Cityscape – the result exceeded all expectations of the team, finding a really well-drawn world of contemporary virtual city.

It should be noted that these algorithms implement only the base form filling virtual space for parts and other things are still needed experts who could independently to Supplement their individual elements and nuances. Besides, a certain part of works on creation of a virtual space is somehow connected with the need to think creatively what neural networks do not always equally effective, unlike humans.

Нейросеть научилась создавать детальные виртуальные города

Нейросеть научилась создавать детальные виртуальные города

According to the materials: