Neural Render Fields NeRF |

The key idea behind NeRF is to represent the 3D shape of an object using a continuous function that maps 3D coordinates to the object's color and reflectance properties. This function is called a "radiance field," and it is learned by a neural network.

Once trained, a NeRF model can be used to generate new views of the object from any viewpoint, as well as estimate the object's 3D shape, surface normals, and albedo. This allows for material properties to dynamically reflect and respond to changing light sources, unlike baked-in textures where for example shadows captured in images can not be easily removed after the fact.

NeRF has been used in various applications such as the 3D reconstruction of objects, indoor and outdoor scenes, and even to generate 3D models of real-world objects that are not in the training data set and can be generated from video data. This is a relatively new process and at the moment has limited output scope but is an evolution of the photogrammetry process driven by AI training.