java opencv gpu解码

转载

mob6454cc6553fc 2024-09-13 16:03:35

文章标签 java opencv gpu解码 opencv 模板匹配 cuda模板匹配 Data 文章分类 Java 后端开发

摘要：

本文主要关注opencv CPU版和cuda版的模板匹配算法，网上cuda版的资料不多，这里做个记录，以后用到也好有个参考。

文章目录

摘要：
1. opencv cuda版配置
2. 源码
3. 结果
4. 总结
5. 2023年1月13日补充测试

1. opencv cuda版配置

opencv cuda版需要自己用cmake编译，编译过程并不复杂，cmake编译成vs的项目，然后用vs编译成opencv_worldXXX.dll.编译过程可参考link1,link2

编译后的opencv 目录
vs项目配置，包含目录，库目录，连接器

注：这里的opencv_world341.lib对应的opencv_world341.dll有两种办法加入到项目中，一是：在系统环境变量中添加它的路径，二是：把它直接复制到项目.exe所在路径。如果有第三种方法,比如在vs项目的什么地方添加一下什么路径就行的话，还请告知，因为我总觉得这两者都不怎么高效，特别是需要移植到不同计算机上时。

2. 源码

包含cpu、gpu版的模板匹配算法demo示例，还有一个速度对比测试。

# include <opencv.hpp>
# include <iostream>
# include <time.h>

int gpuTemplateMatch(const cv::Mat &srcImage, const cv::Mat &tempImage, cv::Mat &result,
	double &matchVal, cv::Point &matchLoc, int mode)
{
	if (srcImage.empty() || tempImage.empty())
	{
		std::cout << "ERROR:In function gpuTemplateMatch: input image is empty! \n";
		return -1;
	}
	cv::cuda::GpuMat d_result;
	cv::cuda::GpuMat d_srcImage;
	cv::cuda::GpuMat d_tempImage;
	d_srcImage.upload(srcImage);
	d_tempImage.upload(tempImage);

	cv::Ptr<cv::cuda::TemplateMatching> alg ;
	switch (mode)
	{
	case 0:
		//R = sum (t-Roi)^2
		alg = cv::cuda::createTemplateMatching(srcImage.type(), cv::TM_SQDIFF);
		alg->match(d_srcImage, d_tempImage, d_result);
		d_result.download(result);
		cv::minMaxLoc(result, &matchVal, NULL, &matchLoc, NULL);
		break;
	case 1:
		//R = sum (t-Roi)^2/(sqrt(sum t^2   *  sum Roi^2))
		alg = cv::cuda::createTemplateMatching(srcImage.type(), cv::TM_SQDIFF_NORMED);
		alg->match(d_srcImage, d_tempImage, d_result);
		d_result.download(result);
		cv::minMaxLoc(result, &matchVal, NULL, &matchLoc, NULL);
		break;
	case 2:
		//R = sum t*Roi
		alg = cv::cuda::createTemplateMatching(srcImage.type(), cv::TM_CCORR);
		alg->match(d_srcImage, d_tempImage, d_result);
		d_result.download(result);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	case 3:
		//R = sum t*Roi / (sqrt(sum t^2   *  sum Roi^2))
		alg = cv::cuda::createTemplateMatching(srcImage.type(), cv::TM_CCORR_NORMED);
		alg->match(d_srcImage, d_tempImage, d_result);
		d_result.download(result);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	case 4:
		//R = sum t1*Roi1
		//t1 = t - t_mean
		//Roi1 = Roi - Roi_mean
		alg = cv::cuda::createTemplateMatching(srcImage.type(), cv::TM_CCOEFF);
		alg->match(d_srcImage, d_tempImage, d_result);
		d_result.download(result);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	case 5:
		//R = sum t1*Roi1 / (sqrt(sum t1^2   *  sum Roi1^2))
		//t1 = t - t_mean
		//Roi1 = Roi - Roi_mean
		alg = cv::cuda::createTemplateMatching(srcImage.type(), cv::TM_CCOEFF_NORMED);
		alg->match(d_srcImage, d_tempImage, d_result);
		d_result.download(result);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	default:
		//cv::matchTemplate(srcImage, tempImage, result, cv::TM_CCOEFF_NORMED);
		alg = cv::cuda::createTemplateMatching(srcImage.type(), cv::TM_CCOEFF_NORMED);
		alg->match(d_srcImage, d_tempImage, d_result);
		d_result.download(result);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	}

	return 0;
}


int cpuTemplateMatch(const cv::Mat &srcImage, const cv::Mat &tempImage, cv::Mat &result,
	double &matchVal, cv::Point &matchLoc, int mode)
{
	//https://docs.opencv.org/3.4.12/de/da9/tutorial_template_matching.html
	if (srcImage.empty() || tempImage.empty())
	{
		std::cout << "ERROR:In function cpuTemplateMatch: input image is empty! \n";
		return -1;
	}

	//cv::Mat result;

	int result_w = srcImage.cols - tempImage.cols;
	int result_h = srcImage.rows - tempImage.rows;
	if (result_w < 0 || result_h < 0)
	{
		std::cout << "ERROR:in function opencvTemplateMatch: roi image's size should be larger than tamplate's \n";
		return -1;
	}
	//result.create(result_h, result_w, CV_32FC1);
	switch (mode)
	{
	case 0:
		//R = sum (t-Roi)^2
		cv::matchTemplate(srcImage, tempImage, result, cv::TM_SQDIFF);
		cv::minMaxLoc(result, &matchVal, NULL, &matchLoc, NULL);
		break;
	case 1:
		//R = sum (t-Roi)^2/(sqrt(sum t^2   *  sum Roi^2))
		cv::matchTemplate(srcImage, tempImage, result, cv::TM_SQDIFF_NORMED);
		cv::minMaxLoc(result, &matchVal, NULL, &matchLoc, NULL);
		break;
	case 2:
		//R = sum t*Roi
		cv::matchTemplate(srcImage, tempImage, result, cv::TM_CCORR);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	case 3:
		//R = sum t*Roi / (sqrt(sum t^2   *  sum Roi^2))
		cv::matchTemplate(srcImage, tempImage, result, cv::TM_CCORR_NORMED);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	case 4:
		//R = sum t1*Roi1
		//t1 = t - t_mean
		//Roi1 = Roi - Roi_mean
		cv::matchTemplate(srcImage, tempImage, result, cv::TM_CCOEFF);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	case 5:
		//R = sum t1*Roi1 / (sqrt(sum t1^2   *  sum Roi1^2))
		//t1 = t - t_mean
		//Roi1 = Roi - Roi_mean
		cv::matchTemplate(srcImage, tempImage, result, cv::TM_CCOEFF_NORMED);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	default:
		cv::matchTemplate(srcImage, tempImage, result, cv::TM_CCOEFF_NORMED);
		cv::minMaxLoc(result, NULL, &matchVal, NULL, &matchLoc);
		break;
	}
	
	return 0;
}


int speedTest()
{
	std::map<int, std::string> matchMode =
	{
		{0,"TM_SQDIFF"},
		{1,"TM_SQDIFF_NORMED"},
		{2,"TM_CCORR"},
		{3,"TM_CCORR_NORMED"},
		{4,"TM_CCOEFF"},
		{5,"cv::TM_CCOEFF_NORMED"}
	};
	/*std::string srcPath = "K:\\imageData\\totalBoard\\image2\\00000103_1-1.png";
	std::string tempPath = "K:\\imageData\\totalBoard\\image2\\00000103_1-2.png";
	cv::Mat srcImage = cv::imread(srcPath, 0);
	cv::Mat tempImage = cv::imread(tempPath, 0);*/

	cv::Mat srcImage = cv::Mat::zeros(cv::Size(200,200),CV_8UC1);
	cv::Mat tempImage = cv::Mat::ones(cv::Size(20, 20), CV_8UC1);
	clock_t start, end;
	double matchVal;
	cv::Point matchLoc;
	cv::Mat result;
	cv::cuda::GpuMat d_srcImage, d_tempImage, d_result;
	cv::Ptr<cv::cuda::TemplateMatching> alg;
	int TIMES = 10;

	for (int mode = 0; mode < 6; mode++)
	{
		for (int size = 100; size < 1000; size += 100)
		{
			//resize the image 
			cv::resize(srcImage, srcImage, cv::Size(2000 , 2000 ));
			cv::resize(tempImage, tempImage, cv::Size(size, size));
			cv::randu(srcImage,cv::Scalar(0),cv::Scalar(255));
			cv::randu(tempImage,cv::Scalar(0),cv::Scalar(255));
			d_srcImage.upload(srcImage);
			d_tempImage.upload(tempImage);
			//gpu match
			start = clock();
			for (int times = 0; times < TIMES; times++)
			{
				gpuTemplateMatch(srcImage, tempImage, result, matchVal, matchLoc, mode);
			}
		   end = clock();
			auto runtime_gpu = (end - start) / TIMES * 1000 / CLOCKS_PER_SEC;

			//cpu match
			start = clock();
			for (int times = 0; times < TIMES; times++)
			{
				cpuTemplateMatch(srcImage, tempImage, result, matchVal, matchLoc, mode);
			}
			end = clock();
			auto runtime_cpu = (end - start) / TIMES * 1000 / CLOCKS_PER_SEC;

			//gpu compute only
			start = clock();
			for (int times = 0; times < TIMES; times++)
			{
				alg = cv::cuda::createTemplateMatching(srcImage.type(), mode);//
				alg->match(d_srcImage, d_tempImage, d_result);
			}
			end = clock();
			auto runtime_gpuComputing = (end - start) / TIMES * 1000 / CLOCKS_PER_SEC;
			
			printf("[+++++++++++++++++++++++++++++++++++++++++++++]\n");
			printf("srcSize=[%d,%d], tempSize=[%d,%d]\n", srcImage.rows, srcImage.cols, tempImage.rows, tempImage.cols);
			printf("match mode:%s\n", matchMode[mode].c_str());
			printf("gpu total runtime:%d ms\n", runtime_gpu);
			printf("cpu total runtime:%d ms\n",runtime_cpu);
			printf("cpuT / gpuT :%3f\n", double(runtime_cpu) / double(runtime_gpu));
			printf("gpu compute time:%dms\n", runtime_gpuComputing);
		}
	}
	return 0;
}


int gpuTemplateMatchDemo()
{
	std::string srcPath = "K:\\imageData\\totalBoard\\image2\\00000103_1-1.png";
	std::string tempPath = "K:\\imageData\\totalBoard\\image2\\00000103_1-2.png";
	cv::Mat srcImage = cv::imread(srcPath, 0);
	cv::Mat tempImage = cv::imread(tempPath, 0);

	//match
	double matchVal;
	cv::Point matchLoc;
	cv::Mat result;
	int mode = 3;
	gpuTemplateMatch(srcImage, tempImage, result, matchVal, matchLoc, mode);

	//show result
	std::cout << "matchVal = " << matchVal << std::endl;
	cv::Point topLeft = matchLoc;
	cv::Point bottomRight = cv::Point(topLeft.x + tempImage.cols, topLeft.y + tempImage.rows);
	cv::Mat drawImage = cv::imread(srcPath);
	cv::rectangle(drawImage, cv::Rect(topLeft, bottomRight), cv::Scalar(0, 255, 0),2);
	cv::imshow("srcImage", srcImage);
	cv::imshow("tempImage", tempImage);
	cv::imshow("drawImage", drawImage);
	//show results
	cv::normalize(result, result, 0, 1, cv::NORM_MINMAX);
	cv::imshow("result", result);
	cv::waitKey(0);
	cv::destroyAllWindows();

	return 0;
}


int cpuTemplateMatchDemo()
{
	//prepare image and template
	std::string srcPath = "K:\\imageData\\totalBoard\\image2\\00000103_1-1.png";
	std::string tempPath = "K:\\imageData\\totalBoard\\image2\\00000103_1-2.png";
	cv::Mat srcImage = cv::imread(srcPath, 0);
	cv::Mat tempImage = cv::imread(tempPath, 0);

	//match
	double matchVal;
	cv::Point matchLoc;
	cv::Mat result;
	int mode = 1;
	cpuTemplateMatch(srcImage, tempImage, result, matchVal, matchLoc, mode);

	//show result
	std::cout << "matchVal = " << matchVal << std::endl;
	cv::Point topLeft = matchLoc;
	cv::Point bottomRight = cv::Point(topLeft.x + tempImage.cols, topLeft.y + tempImage.rows);
	cv::Mat drawImage = cv::imread(srcPath);
	cv::rectangle(drawImage, cv::Rect(topLeft, bottomRight), cv::Scalar(0, 255, 0),2); 
	cv::imshow("srcImage", srcImage);
	cv::imshow("tempImage", tempImage);
	cv::imshow("drawImage", drawImage);
	//show results
	cv::normalize(result, result, 0, 1, cv::NORM_MINMAX);
	cv::imshow("result", result);
	cv::waitKey(0);
	cv::destroyAllWindows();

	return 0;
}


int main()
{
	gpuTemplateMatchDemo();
	cpuTemplateMatchDemo();
	speedTest();
	return 0;
}

3. 结果

demo
速度对比(部分结果)

[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[100,100]
match mode:TM_SQDIFF_NORMED
gpu total runtime:103 ms
cpu total runtime:106 ms
cpuT / gpuT :1.029126
gpu compute time:91ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[200,200]
match mode:TM_SQDIFF_NORMED
gpu total runtime:103 ms
cpu total runtime:95 ms
cpuT / gpuT :0.922330
gpu compute time:90ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[300,300]
match mode:TM_SQDIFF_NORMED
gpu total runtime:101 ms
cpu total runtime:99 ms
cpuT / gpuT :0.980198
gpu compute time:89ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[400,400]
match mode:TM_SQDIFF_NORMED
gpu total runtime:101 ms
cpu total runtime:97 ms
cpuT / gpuT :0.960396
gpu compute time:90ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[500,500]
match mode:TM_SQDIFF_NORMED
gpu total runtime:100 ms
cpu total runtime:94 ms
cpuT / gpuT :0.940000
gpu compute time:90ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[600,600]
match mode:TM_SQDIFF_NORMED
gpu total runtime:111 ms
cpu total runtime:91 ms
cpuT / gpuT :0.819820
gpu compute time:102ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[700,700]
match mode:TM_SQDIFF_NORMED
gpu total runtime:111 ms
cpu total runtime:91 ms
cpuT / gpuT :0.819820
gpu compute time:102ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[800,800]
match mode:TM_SQDIFF_NORMED
gpu total runtime:110 ms
cpu total runtime:89 ms
cpuT / gpuT :0.809091
gpu compute time:102ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[900,900]
match mode:TM_SQDIFF_NORMED
gpu total runtime:108 ms
cpu total runtime:85 ms
cpuT / gpuT :0.787037
gpu compute time:101ms

速度对比测试时的GPU状态

4. 总结

GPU加速模板匹配看起来效果并不是很好，测试了不同大小的图片有的情况速度会超过CPU，本来觉得应该会有几倍的加速效果，但是其实并没有，大多数情况下反而是变慢了。开始觉得是cpu向gpu传图的过程耗时较多，后面去掉传图的过程只看匹配过程，它的计算就是比cpu的慢，不知道是不是因为这块GPU太低端了。

5. 2023年1月13日补充测试

2023年1月13日补充测试

平台

opencv 4.4.0
cuda 11.1
GPU RTX 3090
CPU i9 10900x @3.7GHZ

结论

在用RTX 3090测试opencv模板匹配的时候，对于比较大的图片搜索图2000 * 2000，模板300 * 300～500 * 500之间时GPU版的模板匹配的速度是CPU版的5倍左右，但是在搜索图为200 * 200，模板在20 * 20～60 * 60的情况下，CPU版的速度是GPU版的2倍左右。也就是说当搜索图和模板图都较小的情况下想用GPU加速是没用的

[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[100,100]
match mode:TM_SQDIFF
gpu total runtime:20 ms
cpu total runtime:129 ms
cpuT / gpuT :6.450000
gpu compute time:8ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[300,300]
match mode:TM_SQDIFF
gpu total runtime:20 ms
cpu total runtime:255 ms
cpuT / gpuT :12.750000
gpu compute time:15ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[500,500]
match mode:TM_SQDIFF
gpu total runtime:20 ms
cpu total runtime:150 ms
cpuT / gpuT :7.500000
gpu compute time:10ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[700,700]
match mode:TM_SQDIFF
gpu total runtime:25 ms
cpu total runtime:147 ms
cpuT / gpuT :5.880000
gpu compute time:11ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[900,900]
match mode:TM_SQDIFF
gpu total runtime:24 ms
cpu total runtime:150 ms
cpuT / gpuT :6.250000
gpu compute time:8ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[100,100]
match mode:TM_SQDIFF_NORMED
gpu total runtime:21 ms
cpu total runtime:141 ms
cpuT / gpuT :6.714286
gpu compute time:21ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[300,300]
match mode:TM_SQDIFF_NORMED
gpu total runtime:34 ms
cpu total runtime:272 ms
cpuT / gpuT :8.000000
gpu compute time:27ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[500,500]
match mode:TM_SQDIFF_NORMED
gpu total runtime:18 ms
cpu total runtime:163 ms
cpuT / gpuT :9.055556
gpu compute time:11ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[700,700]
match mode:TM_SQDIFF_NORMED
gpu total runtime:25 ms
cpu total runtime:159 ms
cpuT / gpuT :6.360000
gpu compute time:21ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[900,900]
match mode:TM_SQDIFF_NORMED
gpu total runtime:33 ms
cpu total runtime:149 ms
cpuT / gpuT :4.515152
gpu compute time:10ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[100,100]
match mode:TM_CCORR
gpu total runtime:16 ms
cpu total runtime:88 ms
cpuT / gpuT :5.500000
gpu compute time:9ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[300,300]
match mode:TM_CCORR
gpu total runtime:19 ms
cpu total runtime:225 ms
cpuT / gpuT :11.842105
gpu compute time:10ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[500,500]
match mode:TM_CCORR
gpu total runtime:24 ms
cpu total runtime:125 ms
cpuT / gpuT :5.208333
gpu compute time:19ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[700,700]
match mode:TM_CCORR
gpu total runtime:18 ms
cpu total runtime:122 ms
cpuT / gpuT :6.777778
gpu compute time:3ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[900,900]
match mode:TM_CCORR
gpu total runtime:12 ms
cpu total runtime:126 ms
cpuT / gpuT :10.500000
gpu compute time:5ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[100,100]
match mode:TM_CCORR_NORMED
gpu total runtime:27 ms
cpu total runtime:139 ms
cpuT / gpuT :5.148148
gpu compute time:20ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[300,300]
match mode:TM_CCORR_NORMED
gpu total runtime:24 ms
cpu total runtime:264 ms
cpuT / gpuT :11.000000
gpu compute time:23ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[500,500]
match mode:TM_CCORR_NORMED
gpu total runtime:35 ms
cpu total runtime:161 ms
cpuT / gpuT :4.600000
gpu compute time:22ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[700,700]
match mode:TM_CCORR_NORMED
gpu total runtime:35 ms
cpu total runtime:154 ms
cpuT / gpuT :4.400000
gpu compute time:11ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[900,900]
match mode:TM_CCORR_NORMED
gpu total runtime:29 ms
cpu total runtime:152 ms
cpuT / gpuT :5.241379
gpu compute time:10ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[100,100]
match mode:TM_CCOEFF
gpu total runtime:20 ms
cpu total runtime:115 ms
cpuT / gpuT :5.750000
gpu compute time:16ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[300,300]
match mode:TM_CCOEFF
gpu total runtime:29 ms
cpu total runtime:244 ms
cpuT / gpuT :8.413793
gpu compute time:31ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[500,500]
match mode:TM_CCOEFF
gpu total runtime:32 ms
cpu total runtime:143 ms
cpuT / gpuT :4.468750
gpu compute time:17ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[700,700]
match mode:TM_CCOEFF
gpu total runtime:25 ms
cpu total runtime:138 ms
cpuT / gpuT :5.520000
gpu compute time:8ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[900,900]
match mode:TM_CCOEFF
gpu total runtime:20 ms
cpu total runtime:138 ms
cpuT / gpuT :6.900000
gpu compute time:8ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[100,100]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:27 ms
cpu total runtime:143 ms
cpuT / gpuT :5.296296
gpu compute time:25ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[300,300]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:29 ms
cpu total runtime:275 ms
cpuT / gpuT :9.482759
gpu compute time:43ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[500,500]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:22 ms
cpu total runtime:166 ms
cpuT / gpuT :7.545455
gpu compute time:15ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[700,700]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:30 ms
cpu total runtime:160 ms
cpuT / gpuT :5.333333
gpu compute time:27ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[2000,2000], tempSize=[900,900]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:30 ms
cpu total runtime:155 ms
cpuT / gpuT :5.166667
gpu compute time:13ms

D:\myAPP\vs-proj\cudaTemplateMatch\x64\Release>D:\myAPP\vs-proj\cudaTemplateMatch\x64\Release\cudaTemplateMatch.exe
matchVal = 0.999751
matchVal = 0.00049773
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[10,10]
match mode:TM_SQDIFF
gpu total runtime:0 ms
cpu total runtime:3 ms
cpuT / gpuT :inf
gpu compute time:0ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[30,30]
match mode:TM_SQDIFF
gpu total runtime:5 ms
cpu total runtime:2 ms
cpuT / gpuT :0.400000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[50,50]
match mode:TM_SQDIFF
gpu total runtime:3 ms
cpu total runtime:2 ms
cpuT / gpuT :0.666667
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[70,70]
match mode:TM_SQDIFF
gpu total runtime:4 ms
cpu total runtime:1 ms
cpuT / gpuT :0.250000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[90,90]
match mode:TM_SQDIFF
gpu total runtime:4 ms
cpu total runtime:1 ms
cpuT / gpuT :0.250000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[10,10]
match mode:TM_SQDIFF_NORMED
gpu total runtime:2 ms
cpu total runtime:3 ms
cpuT / gpuT :1.500000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[30,30]
match mode:TM_SQDIFF_NORMED
gpu total runtime:4 ms
cpu total runtime:2 ms
cpuT / gpuT :0.500000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[50,50]
match mode:TM_SQDIFF_NORMED
gpu total runtime:3 ms
cpu total runtime:2 ms
cpuT / gpuT :0.666667
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[70,70]
match mode:TM_SQDIFF_NORMED
gpu total runtime:4 ms
cpu total runtime:2 ms
cpuT / gpuT :0.500000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[90,90]
match mode:TM_SQDIFF_NORMED
gpu total runtime:3 ms
cpu total runtime:2 ms
cpuT / gpuT :0.666667
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[10,10]
match mode:TM_CCORR
gpu total runtime:0 ms
cpu total runtime:2 ms
cpuT / gpuT :inf
gpu compute time:0ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[30,30]
match mode:TM_CCORR
gpu total runtime:2 ms
cpu total runtime:1 ms
cpuT / gpuT :0.500000
gpu compute time:1ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[50,50]
match mode:TM_CCORR
gpu total runtime:2 ms
cpu total runtime:1 ms
cpuT / gpuT :0.500000
gpu compute time:1ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[70,70]
match mode:TM_CCORR
gpu total runtime:2 ms
cpu total runtime:2 ms
cpuT / gpuT :1.000000
gpu compute time:1ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[90,90]
match mode:TM_CCORR
gpu total runtime:2 ms
cpu total runtime:1 ms
cpuT / gpuT :0.500000
gpu compute time:1ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[10,10]
match mode:TM_CCORR_NORMED
gpu total runtime:1 ms
cpu total runtime:2 ms
cpuT / gpuT :2.000000
gpu compute time:1ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[30,30]
match mode:TM_CCORR_NORMED
gpu total runtime:4 ms
cpu total runtime:2 ms
cpuT / gpuT :0.500000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[50,50]
match mode:TM_CCORR_NORMED
gpu total runtime:4 ms
cpu total runtime:2 ms
cpuT / gpuT :0.500000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[70,70]
match mode:TM_CCORR_NORMED
gpu total runtime:5 ms
cpu total runtime:2 ms
cpuT / gpuT :0.400000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[90,90]
match mode:TM_CCORR_NORMED
gpu total runtime:4 ms
cpu total runtime:2 ms
cpuT / gpuT :0.500000
gpu compute time:3ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[10,10]
match mode:TM_CCOEFF
gpu total runtime:2 ms
cpu total runtime:3 ms
cpuT / gpuT :1.500000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[30,30]
match mode:TM_CCOEFF
gpu total runtime:4 ms
cpu total runtime:1 ms
cpuT / gpuT :0.250000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[50,50]
match mode:TM_CCOEFF
gpu total runtime:4 ms
cpu total runtime:2 ms
cpuT / gpuT :0.500000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[70,70]
match mode:TM_CCOEFF
gpu total runtime:4 ms
cpu total runtime:2 ms
cpuT / gpuT :0.500000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[90,90]
match mode:TM_CCOEFF
gpu total runtime:4 ms
cpu total runtime:1 ms
cpuT / gpuT :0.250000
gpu compute time:2ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[10,10]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:3 ms
cpu total runtime:3 ms
cpuT / gpuT :1.000000
gpu compute time:5ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[30,30]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:7 ms
cpu total runtime:2 ms
cpuT / gpuT :0.285714
gpu compute time:4ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[50,50]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:5 ms
cpu total runtime:2 ms
cpuT / gpuT :0.400000
gpu compute time:4ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[70,70]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:5 ms
cpu total runtime:2 ms
cpuT / gpuT :0.400000
gpu compute time:3ms
[+++++++++++++++++++++++++++++++++++++++++++++]
srcSize=[200,200], tempSize=[90,90]
match mode:cv::TM_CCOEFF_NORMED
gpu total runtime:5 ms
cpu total runtime:2 ms
cpuT / gpuT :0.400000
gpu compute time:4ms

D:\myAPP\vs-proj\cudaTemplateMatch\x64\Release>

本文章为转载内容，我们尊重原作者对文章享有的著作权。如有内容错误或侵权问题，欢迎原作者联系我们进行内容更正或删除文章。