OpenCV的SGBM with GPU和without GPU的性能对比

测试机器为AMD5800H+3050Ti

测试代码：

#include 
#include 
#include 
#include 
#include 
#include 
#include
#include 

using namespace std;

void calcu_disparity(cv::Mat& left, cv::Mat& right, cv::Mat& disparity)
{
    auto t_a = chrono::high_resolution_clock::now();
    cv::Ptr sgbm = cv::StereoSGBM::create(
        0, 96, 9, 8 * 9 * 9, 32 * 9 * 9, 1, 63, 10, 100, 32);
    cv::Mat disparity_sgbm;
    sgbm->compute(left, right, disparity_sgbm);
    disparity_sgbm.convertTo(disparity, CV_32F, 1.0 / 16.0f);
    auto t_b = chrono::high_resolution_clock::now();
    cout<<"no cuda:"<< chrono::duration_cast(t_b-t_a).count()<<"微秒"< sgbm = cv::cuda::createStereoSGM(
        0, 128, 8, 180, 3, cv::cuda::StereoSGM::MODE_HH);
    cv::Mat disparity_sgbm(left.size(), CV_16S);
    cv::cuda::GpuMat cudaDisparityMap(left.size(), CV_16S);
    cv::cuda::GpuMat cudaDrawColorDisparity(left.size(), CV_8UC4);
    cv::cuda::GpuMat cudaLeftFrame, cudaRightFrame;
    cudaLeftFrame.upload(left);
    cudaRightFrame.upload(right);
    sgbm->compute(cudaLeftFrame, cudaRightFrame, cudaDisparityMap);
    cudaDisparityMap.download(disparity_sgbm);
    disparity_sgbm.convertTo(disparity, CV_32F, 1.0 / 16.0f);
    auto t_b = chrono::high_resolution_clock::now();
    cout<<"with cuda:"<< chrono::duration_cast(t_b-t_a).count()<<"微秒"< 
注意，参数还没调，不知道影响有多大。
 
结果：
 
 
粗算一下，应该是13-15倍的速度提升
 
视差和点云图 
without GPU:
 
 
 
 
with GPU: