NeRF
A neural network that synthesizes novel views of 3D scenes from a set of 2D photographs.
Overview
Neural Radiance Fields (NeRF), introduced in 2020, represent 3D scenes as continuous functions mapping 3D coordinates and viewing directions to color and density values, parameterized by a neural network. Given a set of 2D photographs of a scene, NeRF learns to render photorealistic novel views from any angle.
Key Details
NeRF works by casting rays through each pixel and querying the neural network along each ray using volume rendering. Extensions like Instant-NGP (using hash grids) dramatically accelerated training. NeRF has transformed 3D content creation, enabling applications in virtual reality, e-commerce product visualization, heritage preservation, and film production.