What’s the difference between a point cloud and a Gaussian splat?

In the world of LiDAR and 3D scanning, point clouds come up in conversation on a regular basis. Point clouds exist entirely in the virtual world, a digital representation of the data gathered by a 3D scanner or other tool equipped with a LiDAR sensor.

So what, exactly, is a point cloud?

What is a point cloud?

A point cloud is a visualization of data collected by a LiDAR scanner and plotted in 3D space. Each point in the point cloud is correlated to the xyz coordinates of a specific location in real space, such as the surface of an object. 

A point cloud may represent a single object or a large space, such as a stockpile or the interior of a warehouse. Point clouds are excellent for capturing and displaying the geometry of a shape or space, making them a powerful tool for applications requiring pinpoint accuracy.

What are point clouds used for?

The applications for point clouds are virtually unlimited. Surveyors, engineers, and architects may use point clouds to create virtual models of a space or component. A museum may create 3D renderings of historical artifacts using a point cloud. The entertainment industry may use a point cloud as a starting point for the creation of a virtual set. The applications for point clouds are as numerous as the applications for 3D scanners.

What is a Gaussian Splat?

Another term often heard in the world of 3D rendering is Gaussian splatting. A Gaussian splat is a rasterization technique that creates 3D renderings from a collection of 2D images. A single gaussian—or a single point in the splat—describes the position, scale, color, and transparency of the corresponding point in the source images. These all come together to create a photorealistic image.

To be clear, this is a highly simplified explanation of a Gaussian splat. Gaussian splatting is a complex and rapidly developing technology.

What is the difference between a point cloud and a Gaussian Splat?

There are many differences between point clouds and Gaussian splats. In very simple terms, a point cloud converts data into a series of points that are identical in shape and size. Point clouds can convey some information about color, but the real strength of a point cloud is the capture of geometry.

A Gaussian splat does not convert data in the same way; instead, a Gaussian splat is a collection of colored shapes. The complexity of a Gaussian splat creates a more photorealistic final image because each point conveys more information about its corresponding location in the source image or material.

The processing of a Gaussian splat is more complex as well. Gaussian splatting may employ complex algorithms and rasterizers to process a final image, and this process often leverages artificial intelligence (AI).

How to create a point cloud.

The first step to creating an accurate point cloud is investing in a high-quality 3D laser scanner. E38 Survey Solutions is proud to offer the Lixel Kity K1 and the Lixel L2 Pro, from XGRIDS. Both of these 3D scanners combine a user-friendly interface with powerful LiDAR technology to capture high-quality data for processing into accurate point clouds for a wide variety of applications.

Learn more >