The RKD forest is a project we created. For LSH, we used E²LSH (from MIT) and for the BBD tree we used ANN library. ANN was easier to handle than E²LSH. We used SIFT data for the measurements.
BBD tree was the most accurate of all, while RKD forest the fastest of all. BBD couldn’t decrease much of its execution time by approximation parameters (such as epsilon) will E²LSH could, resulting in faster execution with just a bit loss in precision (in respect with BBD tree).