2-dimensional KDtree with cuda, only NN (Nearest Neighbour) search is implemented. The tree nodes structure is built on CPU and transferred to GPU before used.
On laptop with Intel Core i7-9750H and NVIDIA GeForce GTX 1650:
# build kdtree with 2k sample points
----------------------------------------------------------------
Benchmark Time CPU Iterations
----------------------------------------------------------------
BM_create_kdtree_ref 259041 ns 259010 ns 2696
BM_create_kdtree 222636 ns 222629 ns 3126
BM_create_kdtree_cuda 228376 ns 228338 ns 3067
# NN search with 2k sample points and 45k queries
-----------------------------------------------------------
Benchmark Time CPU Iterations
-----------------------------------------------------------
BM_nnsearch_ref 9337326 ns 9336433 ns 72
BM_nnsearch 8687874 ns 8687220 ns 79
BM_nnsearch_cuda 375813 ns 375790 ns 1678- Samples need OpenCV
- Tests need Google Test
- Benchmarks need Google Benchmark
- kd-tree (also the ref in tests and benchmarks)
- KD-TREE ON GPU FOR 3D POINT SEARCHING