However, I can tell you from my experience of using it for alignment in my
particular data, that doing a prior transformation of the input cloud (very
coarse alignment by normals) had a very positive effect on the performance
of the following step (alignment using FPFH).
I wouldn't go as far as to say it invalidates the features, but I think
rec-computing the features can help to finetune them.
> If the features(FPFHSignature33) of a point cloud was computed. Then, the
> point cloud was transformed using pcl::TransformPointCloud.
> Does this invalidate the computed features? Should I re-compute them again
> for the transformed point cloud?