Point Cloud

A point cloud is a way of representing three-dimensional space as a collection of very small points.

Think of each grain of sand carrying its own position and color information. When tens or hundreds of thousands of these points come together, they look to the human eye like a smooth, solid space. It works on a similar principle to pointillism: when tiny dots are packed densely enough, they form a picture. The denser and more numerous the points, the more three-dimensional the result looks; the sparser they are, the rougher and more sketch-like it appears.

AI models that reconstruct real spaces in 3D from just a few photos break down every surface in the photos into points, calculate each point's position and color, and then scatter these points in space to build a 3D scene. The result can then be examined by rotating and moving around it, as if moving a camera through the space.

In the article, it's mentioned that the 3D scenes generated by World Labs' world model Atlas support "point clouds and 3D Gaussian splats" as output formats. One thing that's easy to misunderstand is that a point cloud isn't a specific company's technology—it's just one of several ways to represent 3D space.

On a smartphone with a LiDAR sensor (such as an iPhone Pro model), open a 3D scanning app and slowly sweep it around a room. As dots appear on screen one by one, the shape of the space gradually emerges—that's the point cloud being built in real time.