NVIDIA has introduced its new technology called Instant NeRF, which can capture entire 3D scenes using just a few 2D photos. The demonstration was made using four images from different angles of a single scene. In just a few moments, the software can generate a 3D environment that can then be explored from any angle. Using an AI model, the software approximates what is at certain angles not covered by 2D images.
NVIDIA’s NeRF can help you develop video games and autonomous driving technologies
NeRF is short for Neural Radiance Field, a technology that trains artificial intelligence to make 3D objects using 2D photography. The technology could be used in areas such as video games or self-driving cars, using the information the cameras see to approximate the distance between vehicles in traffic. At the same time, the technology could be used to train industrial robots.
The advantage of NeRF over other technologies of this kind in the past is that the “training” to generate 3D images is done in just a few seconds, and then be able to generate new media and 3D models based on photos in just a few moments.
NVIDIA says that NeRF works mainly with static subjects, and it is preferable that the images do not contain too many differences in motion when inserted into the system. Otherwise, some elements may appear “blurred”, as in a long-exposure photograph in which subjects move.
To run as fast as possible, NVIDIA recommends using NeRF on video cards equipped with Tensor cores, such as those in the RTX range, but can also be used on video cards without native AI acceleration. Of course, performance will be poorer without the presence of Tensor cores.
source: Engadget