Closed Sheradil closed 1 year ago
Ok, so the documentation is weird. I fixed it in the end. The documentation says: max_correspondence_distance (float) – Maximum correspondence points-pair distance. In their tutorial this parameter is called "threshold" and I would expect the algorithm to run until the error is less than the threshold. However, the algorithm doesn't start if the error is bigger than the threshold. This should be formulated more precisely in the documentation. And especially in the tutorial this has to be fixed.
Checklist
master
branch).Describe the issue
I have been trying to run the tutorial of the point cloud registration for a point cloud of my own. It didn't work. I therefore created my own point cloud of 12 points, a rectangle of 3x4 points lying on the X-Y plane. I duplicated this cloud and moved it along the Z axis by 1.0 or by 0.1. For both (modified) point clouds I tried to do ICP and match them with the original cloud. The 'initial guess' transformation matrix is the identity matrix. Even though the clouds obviously do not match perfectly, the error that the RMSE gives is 0. But in both cases, the result of the ICP is the input transformation (so the identity matrix). So basically the ICP is doing nothing. (I tried some other things to address certain things that could go wrong (in my opinion), but nothing worked and this is a quite simple example. I probably did something stupid and just can't see the forest for the trees)
Steps to reproduce the bug
Error message
No error message, but the result is obviously wrong.
Expected behavior
Output the correct transformation matrix.
Open3D, Python and System information
Additional information
No response