The eigen vectors were serialized row major instead of column major, which shouldn't be the case since everything else is ordered column major in libpm. So, it should be assumed that it's also the case for the eigen vectors. This PR will have an impact the users using the following filters:
The eigen vectors were serialized row major instead of column major, which shouldn't be the case since everything else is ordered column major in
libpm
. So, it should be assumed that it's also the case for the eigen vectors. This PR will have an impact the users using the following filters:Ellipsoids (1)
,Ellipsoids (2)
Gestalt (1)
,Gestalt (2)
SamplingSurfaceNormal
SurfaceNormal
and the following descriptors:
keepEigenVectors
keepCovariances
So, retrieving the eigen vectors from a 3D point cloud before this PR could be done like this:
And after this PR it could be done like this:
Then, reconstructing the matrix is the same for both method:
As we can see, it is more straightforward to retrieve the eigen vectors when they ordered column major.