While GTC 2022 revealed many technologies that automatically generate images with artificial intelligence, NVIDIA also introduced a tool set called “GET3D”, which allows game creators or metaverse content builders to quickly set up a large number of different 3D images. object.
In game scenes, metaverses, or virtual visual application environments, for the richness and fidelity of the picture, it is usually necessary to create many 3D objects with different shapes, and try to avoid copying a large number of the same scene objects, but this will require Spend a lot of time modeling.
Therefore, the method of randomly generating 3D objects by random numbers began to appear, but there are still similar or unreasonable compositions. Therefore, the “GET3D” tool set proposed by NVIDIA is to complete deep learning training in advance. The artificial intelligence module can automatically generate 3D objects that are realistic, reasonable and avoid obvious repetition according to the needs of the scene.
NVIDIA said that the “GET3D” tool set can automatically generate 20 objects per second on average. It uses a large number of A100 GPUs to cooperate with 1 million 2D plane image content for deep learning, thereby generating real and vivid 3D images. It can correspond to the needs of various file formats, so that it can be easily applied to videos, games and other content.
In conjunction with the StyleGAN-NADA tool proposed by the NVIDIA research institute, the 3D objects generated by the “GET3D” tool group can make the surface texture map more natural and realistic, making the appearance of the generated 3D objects more visually convincing.