There are several ways to generate Point Normal information, and some work better than others, and some are very much work in progress.
The Point Normal Generation operator is supposed to generate normals just from neighboring information. However it is not quite ready for prime time, and is incredibly slow. It works best with aerial scans of terrain where most of the data is pointing mostly up anyway. If you try to use it on arbitrary clouds like a car or a building, it will have hard time figuring out “inside” and “outside”, so while the normals might be mostly correct, they might be pointing exactly in the opposite direction… Improving this operator is on our To Do list. You can see a tutorial describing the steps of using it here:
docs.thinkboxsoftware.com/produc … rrain.html
The Normal From Scanner Position operator uses the Scanner Position data (if any) to determine the “visibility vector” of each point - this is not the normal of the surface, but the vector connecting the point sample to the scanner’s position. It is good enough for determining the “front” and “back”. If a point cloud connects multiple Scanner Positions, each particle will use the value based on the ScannerIndex channel which describes which scanner took that sample.
docs.thinkboxsoftware.com/produc … erpos.html
If you don’t have a Scanner Position, you could still make it work by creating a reference point approximately where the scanner was placed, and picking it as the reference point in the Set Scanner Position operator (new in v1.1). Then you can add a Normal From Scanner Position below it on the Operators list, and it will use the manually specified position as the target of the visibility vectors. This will only work in limited number of cases where a single scanner was used to create the point cloud, and its position is obvious from the “empty circle” below the tripod usually seen in the point cloud…
Hope this helps!