NVIDIA told regarding the artificial intelligence model GET3D, which will facilitate the work of the creators of virtual 3D-worlds. The algorithm quickly generates people, buildings, vehicles, and many other types of 3D objects. One GPU, the company said, allows GET3D to generate up to 20 objects per second.
Image source: blogs.nvidia.com
NVIDIA engineers trained AI on flat projections of three-dimensional objects made from different angles. Training was done on A100 accelerators – it took only two days to load a million samples into GET3D. The system generates objects in the format of a standard triangular polygon mesh with high-resolution textures. The technology is aimed at creators of 3D worlds, so you can export objects in formats that are compatible with game engines, 3D modeling and movie rendering tools. This can really make it easier for developers to create games and applications of the metaverse – the system will also come in handy in robotics and architecture.
The algorithm distinguishes nuances between similar objects: from vehicles it can generate sedans, trucks, racing cars and vans, from the category of animals it creates images of foxes, rhinos, horses and bears – the set expands as you learn. NVIDIA also talked regarding the StyleGAN-NADA auxiliary AI model, which allows you to change the styles of objects: make a car burnt, turn a building into a haunted house, or give tiger color to any animal. The company is going to train future versions of the GET3D system on real objects, and not on artificially created models.
If you notice an error, select it with the mouse and press CTRL + ENTER.
