Definition
3D Gaussian Splatting
3D Gaussian Splatting is a real-time 3D scene representation technique that models a scene as a collection of 3D Gaussian functions (ellipsoidal splats) fitted to data captured from images or video. Each splat encodes spatial position, orientation, scale, opacity, and view-dependent color via spherical harmonics. By rasterizing these Gaussians onto the image plane, the technique enables high-quality novel view synthesis at real-time rendering speeds, making it a powerful alternative to implicit neural representations like NeRF.
3D Gaussian Splatting works by fitting a large set of 3D Gaussian functions — often called ellipsoidal splats — to a scene reconstructed from multi-view images or video. Each splat encodes a set of properties: position in 3D space, orientation, scale along each axis, opacity, and view-dependent color represented through spherical harmonics. During rendering, these Gaussians are projected (“splatted”) onto the image plane via differentiable rasterization, allowing the full scene to be rendered in a single forward pass without ray marching.
Compared to neural radiance fields (NeRF), 3D Gaussian Splatting offers substantial practical advantages. Rendering is dramatically faster — achieving 30 to over 100 frames per second in real time, versus NeRF’s slow per-pixel ray marching that can take seconds per frame. The representation is also explicit rather than implicit, meaning individual splats can be inspected, edited, or manipulated directly, which simplifies downstream tasks like scene editing and object removal. Visual quality is competitive with or superior to NeRF across many scene types.
These properties make 3D Gaussian Splatting well-suited for a wide range of applications. It is used in augmented and virtual reality, spatial computing platforms, virtual staging for real estate, digital twin creation, and immersive data visualization. The technique is also gaining traction in game and film production pipelines, where real-time photorealistic rendering of captured environments is increasingly valuable.