c++調用tensorflow的pb模型

#include<iostream>
#include<fstream>
#include<opencv2/opencv.hpp>
#include<opencv2/dnn.hpp>

#include <Windows.h>
#pragma comment( lib, "winmm.lib")

using namespace std;
using namespace cv;
using namespace cv::dnn;

Mat predict(Net net, char * dir)
{
	Mat img = imread(dir, -1);
	Mat inputBlob = blobFromImage(img, 1.0 / 255.0, Size(340, 340), Scalar(), false, false);

	net.setInput(inputBlob);

	Mat detection = net.forward();

	return detection;
}

	
int main()
{
	string strResult, strTemp;
	char czPath[256] = { 0 };
	GetModuleFileName(NULL, czPath, 256);  //獲取全路徑
	string strPath = czPath;
	cout << "文件名(帶全路徑) : " << strPath.c_str() << endl;
	int nPos = strPath.find('x'); //查找最後一個 \ 出現的位置
	strTemp = strPath.substr(0, nPos);
	String weight = strTemp+"tmp\\new_tensor_model.pb";
	cout<< weight<< endl;

	Net net = readNetFromTensorflow(weight);

	//設置計算後臺（GPU調用，如需修改按ctrl+點擊setPreferableTarget）
	net.setPreferableBackend(DNN_BACKEND_OPENCV);
	net.setPreferableTarget(DNN_TARGET_OPENCL);

	/*打出節點
    vector<string> outLayersNames = net.getLayerNames();
	int lens = outLayersNames.size();
	cout << lens << endl;
	for(int i = 0; i < lens; i++)
	{ 
		cout << outLayersNames[i] << endl;
	}*/


	DWORD t1, t2;
	char dir[] = "D:/hyd/silk_baby2/si/2/l2_9.jpg";
	Mat detection1 = predict(net, dir);
	t1 = GetTickCount();
	Mat detection = predict(net, dir);
	cout << detection << endl;
	t2 = GetTickCount();
	cout<< (t2 - t1)*1.0 / 1000 << endl;
}