How is this partitioning done automatically? Is it possible to use something like k-means clustering on the centroids and select the primitives for each bounding box that way?