1- Using squared distance to save from calling sqrt all the time (3% faster on my benchmark, not much, but why not)
2- search params for approximate closest point to speed up the algorithm. Up to 2 orders of magnitude faster depending on the test case. In my testing I use this within an ICP algorithm, and the approximation (using far_node_inspect_budget=0) has minimal impact on the final accuracy (less than 1%). With the default settings, the algorithm is exact.
1- Using squared distance to save from calling sqrt all the time (3% faster on my benchmark, not much, but why not) 2- search params for approximate closest point to speed up the algorithm. Up to 2 orders of magnitude faster depending on the test case. In my testing I use this within an ICP algorithm, and the approximation (using far_node_inspect_budget=0) has minimal impact on the final accuracy (less than 1%). With the default settings, the algorithm is exact.