Neural Radiance Fields

Neural Radiance Fields (NeRF) is a method that achieves state-of-the-art results for synthesizing novel views of complex scenes. It uses a fully-connected deep network to represent the scene, whose input is a continuous 5D coordinate (spatial location and viewing direction) and whose output is the volume density and view-dependent emitted radiance at that spatial location. The NeRF is trained to reproduce input images of the scene from known camera poses. By querying the network at locations along camera rays, it renders novel views by accumulating the predicted colors and densities into an image.