lightweight kd tree implementation on top of Magnum/Corrade.
It does not copy you pointcloud, instead you pass (a potentially strided) view of your data to the tree.
In my measurements, constructing the kd tree is about 15 percent faster than nanoflann and a nearest neighbor query is about 10 percent faster.