CloudCompare forum

Hello,
I'm having some trouble understanding a georeferencing function in my point cloud software. Here's my workflow: I use surveyed ground control points with known coordinates to georeference a point cloud. After transforming and aligning the point cloud to these points, the software asks whether I want to "apply global transformation to reference points as well."
I find this question confusing. I simply want a precisely georeferenced point cloud based on my surveyed points. Typically, I select "NO—don't shift reference points." However, I once discovered my point cloud was approximately one meter off from its true position. When I reran the process and selected "YES" instead, the results were correct.
Despite this, the logic escapes me. I understand the point cloud may be slightly shifted upon import, but I don't grasp what this secondary transformation step actually does or why it matters.
Could someone clarify the difference between these two options and when each should be used?
Thank you!

Yes, it can be confusing.

What happens is that if you georeference a point cloud with GCPs that have very large coordinates, all the aligned cloud coordinates will become very large. In which case you would lose numerical accuracy if the coordinates are kept as is, as CloudCompare stores them on 32bits (to save memory). So to avoid this, CloudCompare suggests to 'apply a global shift' to these large coordinates right away, just as what would happen if you try to load a file with large coordinates. This way, the coordinates of the cloud will stay small for the cloud as long as it's loaded in CloudCompare. But they will be properly 'expanded' at export time.

And CloudCompare gives the user the choice in case the final coordinates are not that large, or the user doesn't care too much about the numerical accuracy (even though the general recommendation is clearly to accept the temporary 'Global shift').