Definition
Gaussian splatting is a modern technique for representing and displaying 3D scenes using many small, soft blobs of colour and light called Gaussians. Instead of building a scene from hard surfaces or points, it fills space with these overlapping blobs to recreate how a place really looks.
The approach produces highly realistic, smooth 3D views that can be explored from any angle, and it is fast to display. It has become popular for turning ordinary photos and videos into immersive 3D scenes.
In simple terms, a Gaussian splat is a cloud of tiny coloured blobs that together form a lifelike, viewable 3D scene.
How It Works
Gaussian splatting begins with a set of images of a scene taken from different viewpoints. An algorithm works out the 3D positions, sizes, colours, and transparency of millions of Gaussians so that, when viewed together, they reproduce the original scene from any angle.
Because each Gaussian is a smooth, semi-transparent blob rather than a solid point or triangle, the rendered result looks soft and natural, capturing fine detail and lighting. The representation is also efficient to display in real time, which sets it apart from some traditional 3D methods.
Applications
Gaussian splatting is used in visualisation, virtual tours, gaming, film, and increasingly in reality capture. It offers a way to present captured environments as photorealistic, navigable 3D scenes, complementing more measurement-focused outputs like point clouds and meshes.
Benefits
- Highly realistic, photorealistic 3D scenes.
- Smooth appearance with fine detail and lighting.
- Fast, real-time rendering and exploration.
- Created from ordinary images or video.
- An engaging way to present captured spaces.
Related Terms
- Point Cloud
- Photogrammetry
- 3D Model
- Reality Capture
Frequently Asked Questions
What is Gaussian splatting used for?
It is used to create and display photorealistic, explorable 3D scenes for visualisation, virtual tours, and reality capture.
How is Gaussian splatting different from a point cloud?
A point cloud is made of discrete measured points, while Gaussian splatting uses soft, semi-transparent blobs that blend into a smooth, lifelike image.
What input does Gaussian splatting need?
It is generated from a set of images or video of a scene captured from multiple viewpoints.
Is Gaussian splatting good for measurement?
It excels at realistic visualisation, while point clouds and meshes remain the usual choice when precise measurement is the priority.
About Track3D
Track3D helps teams turn captured imagery and scans into immersive, actionable reality capture experiences. Learn more at Track3D.
