Farthest point sampling algorithm to implement PointNet++

[codejaeger]

Thanks for open-sourcing a TensorFlow package to implement 3D graphics neural layers. I was trying to implement PointNet++ using the existing API and noticed that a sampling algorithm is missing. To be specific, the implementation of PointNet++ requires a version called farthest point sampling. Is there a plan for such an algorithm to be implemented in the future or are contributions welcome for it?