NeRF

NeRF (Neural Radiance Fields) is a deep learning method that models a continuous 5D volumetric scene function to synthesize photorealistic, novel views from a sparse set of 2D input images.

NeRF revolutionizes view synthesis: it represents a 3D scene as a continuous volumetric function, implicitly encoded by a Multilayer Perceptron (MLP). The network takes a 5D coordinate input (3D spatial location $\mathbf{(x, y, z)}$ and 2D viewing direction $\mathbf{(\theta, \phi)}$), outputting volume density $\mathbf{(\sigma)}$ and view-dependent RGB color. This function is optimized against a sparse set of input images with known camera poses. By using differentiable volume rendering, NeRF generates high-fidelity novel views, capturing complex effects like reflections and transparency that challenge traditional 3D modeling.

https://www.matthewtancik.com/nerf