The running time reported in the paper is from C++ implementation. This Matlab version is a re- implementation, and is for the ease of understanding the algorithm. This code is not optimized, and the speed is not representative. The result can be slightly different from the paper due to transferring across platforms.