Are point clouds actually used in practical applications? How can we determine where points should be for an arbitrary surface, such as the one on the right? It seems easy to randomly sample points following some explicit formula, but when we have a complex shape it seems unrealistic to manually input all the points. Also how would we make sure that the entire surface is colored by at least one point? Do we do interpolation between the colors of neighboring points on the surface?
@acpatel Point clouds typically arise from direct measurements of real physical geometry (depth sensors, laser scanners, LiDar, multiview stereo, etc.). So, if you can process or answer queries about point clouds directly this can be desirable, since you don’t have to convert to a different representation. For instance, generating a high-quality mesh fron a point cloud can be quite challenging.
Representing a 3D model in terms of points feels a lot like the undersampling problem we had before: the more points we have in our point cloud, the more accurate our representation, until at a near infinite number of points we get completely continuous surfaces.
@Sleepyhead08 Yes indeed! All the basic issues of sampling definitely arise when measuring geometry.