IRadianceField<T>

Interfaces · AiDotNet.NeuralRadianceFields.Interfaces

Defines the core functionality for neural radiance field models.

Implements: INeuralNetwork<T>

Known implementations: GaussianSplatting<T>, InstantNGP<T>, NeRF<T>

How It Works

For Beginners: A neural radiance field represents a 3D scene as a continuous function.

Traditional 3D representations:

  • Meshes: Surfaces made of triangles
  • Voxels: 3D grid of cubes (like 3D pixels)
  • Point clouds: Collection of 3D points

Neural Radiance Fields (NeRF):

  • Represents scene as a neural network
  • Input: 3D position (X, Y, Z) and viewing direction
  • Output: Color (RGB) and density (opacity/volume density)

Think of it like this:

  • The neural network “knows” what the scene looks like from any position
  • Ask “What’s at position (x, y, z) when viewed from direction (θ, φ)?”
  • Network responds “Color is (r, g, b) and density is σ”

Why this is powerful:

  • Continuous representation (query any position, not limited to discrete grid)
  • View-dependent effects (reflections, specularities)
  • Compact storage (just network weights, not millions of voxels)
  • Novel view synthesis (render from any camera angle)

Applications:

  • Virtual reality and AR: Create photorealistic 3D scenes
  • Film and gaming: Capture real locations and render from any angle
  • Robotics: Build 3D maps of environments
  • Cultural heritage: Digitally preserve historical sites

Methods

Method Summary
Tensor<T> RenderImage(Vector<T> cameraPosition, Matrix<T> cameraRotation, int imageWidth, int imageHeight, T focalLength) Renders an image from a specific camera position and orientation.
Tensor<T> RenderRays(Tensor<T> rayOrigins, Tensor<T> rayDirections, int numSamples, T nearBound, T farBound) Renders rays through the radiance field using volume rendering.
ValueTuple<Tensor<T>, Tensor<T>> QueryField(Tensor<T> positions, Tensor<T> viewingDirections) Queries the radiance field at specific 3D positions and viewing directions.