I am working on implementing a 3d detection architecture, Lasernet, in Tensorflow. The step I\'m stuck on involves taking per-point bounding box predictions from a range image,
