5. Gaussian Splatting

Explicit Differentiable Primitives

This module covers the state-of-the-art alternative to implicit fields: 3D Gaussian Splatting (3DGS). This approach utilizes a set of anisotropic Gaussians that are projected and rasterized onto the screen.

Feature NeRF 3D Gaussian Splatting
Representation Implicit (MLP) Explicit (3D Gaussians)
Core Algorithm Volumetric Ray Marching Differentiable Rasterization
Rendering Speed Low (Seconds per frame) High (Real-time / >100 FPS)
Data Density Continuous Fields Spatially Anchored Primitives

Learners implement the splatting algorithm, handling the 3D-to-2D projection and sorted alpha-blending for high-speed synthesis.

Sandbox: The pre-compiled assets have 7:1 compression of material properties into neural features.