Neural Radiance Fields

Prompt

Identify

You see a [photorealistic, three-dimensional scene] rendered from new camera angles. The view includes [complex lighting, reflections, and translucent materials] that shift naturally as you move.

Principle

A neural network¹ analyzes multiple photographs to learn how light travels through a space. It [calculates the color and density] of light at every invisible point, generating new views.

Example

Ben Mildenhall et al.: "NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis", 2020.

Try it out

Look at a shiny object from different angles and notice how the reflections move; NeRF accurately simulates this exact shifting behavior.

Background

Researchers at UC Berkeley, Google Research, and UC San Diego introduced NeRF in 2020. Before this, creating 3D computer graphics required building complex geometric meshes. NeRF took a radically different approach. Instead of polygons, it uses a massive mathematical formula—a neural network—to map a scene's volume. By feeding the network dozens of standard photographs along with their exact camera positions, the system learns the underlying light field. It deduces how dense objects are and how they scatter light. The result is a continuous volumetric scene² that can render entirely new, highly realistic viewpoints.

Distinction

Unlike traditional photogrammetry, which creates a hollow 3D mesh wrapped in flat textures, NeRF calculates light and volume mathematically. This allows NeRF to perfectly recreate mirrors, glass, and complex shadows that confuse mesh-based systems.

Aftermath

It teaches that deep learning can translate 2D images into continuous volumetric light fields.

Question