将mnist训练的caffemodel生成动态链接库DLL
在项目程序中经常看到动态链接库，非常好奇，想自己实现一下，于是乎尝试一波。就因为这种好奇，每天都被bug所困扰。。。
1. 训练caffemodel
在windows环境下搭建caffe无果，转投Ubuntu。。。
用的caffe--example--mnist中的文件，新建文件夹的话注意改路径，下面为train.sh
#!/usr/bin/env sh set -e /home/fish/caffe/build/tools/caffe train --solver=/home/fish/STUDY/lenet_solver.prototxt
训练好后把lenet_train_test.prototxt和训练好的模型lenet_iter_10000.caffemodel拿出来。
2. 使用cv::dnn里的API加载model，输入图片，进行测试（可跳过）
根据文章https://blog.csdn.net/sushiqian/article/details/78555891，修改模型文件。若图片为白底黑字，bitwise_not一下。
#include #include <opencv2/opencv.hpp> #include <opencv2/dnn.hpp> using namespace std; using namespace cv; using namespace cv::dnn; /* Find best class for the blob (i. e. class with maximal probability) */ static void getMaxClass(const Mat& probBlob, int* classId, double* classProb) { Mat probMat = probBlob.reshape(1, 1); Point classNumber; minMaxLoc(probMat, NULL, classProb, NULL, &classNumber); *classId = classNumber.x; } int main(int argc, char* argv[]) { string modelTxt = "C:\\Users\\ATWER\\Desktop\\lenet_train_test.prototxt"; string modelBin = "C:\\Users\\ATWER\\Desktop\\lenet_iter_10000.caffemodel"; string imgFileName = "C:\\Users\\ATWER\\Desktop\\9.png"; //read image Mat imgSrc = imread(imgFileName); if (imgSrc.empty()) { cout << "Failed to read image " << imgFileName << endl; exit(-1); } Mat img; cvtColor(imgSrc, img, COLOR_BGR2GRAY); //LeNet accepts 28*28 gray image resize(img, img, Size(28, 28)); bitwise_not(img, img); img /= 255; //transfer image(1*28*28) to blob data with 4 dimensions(1*1*28*28) Mat inputBlob = dnn::blobFromImage(img); dnn::Net net; try { net = dnn::readNetFromCaffe(modelTxt, modelBin); } catch (cv::Exception& ee) { cerr << "Exception: " << ee.what() << endl; if (net.empty()) { cout << "Can't load the network by using the flowing files:" << endl; cout << "modelTxt: " << modelTxt << endl; cout << "modelBin: " << modelBin << endl; exit(-1); } } Mat pred; net.setInput(inputBlob, "data");//set the network input, "data" is the name of the input layer pred = net.forward("prob");//compute output, "prob" is the name of the output layer cout << pred << endl; int classId; double classProb; getMaxClass(pred, &classId, &classProb); cout << "Best Class: " << classId << endl; cout << "Probability: " << classProb * 100 << "%" << endl; }
3. 创建动态链接库
参考https://blog.csdn.net/qq_30139555/article/details/103621955
class.h
#include #include <opencv2/opencv.hpp> #include <opencv2/dnn/dnn.hpp> using namespace std; using namespace cv; using namespace cv::dnn; extern "C" _declspec(dllexport) void Classfication(char* imgpath, char* result);
在此处卡的最久，原本我写的是Classfication(string imgpath, string result)，生成dll时没问题，调用时总是System.AccessViolationException: 尝试读取或写入受保护的内存。后来发现要写成指针的形式。
class.cpp
#include #include <opencv2/opencv.hpp> #include <opencv2/dnn/dnn.hpp> #include "class.h" using namespace std; using namespace cv; using namespace cv::dnn; /* Find best class for the blob (i. e. class with maximal probability) */ static void getMaxClass(const Mat& probBlob, int* classId, double* classProb) { Mat probMat = probBlob.reshape(1, 1); Point classNumber; minMaxLoc(probMat, NULL, classProb, NULL, &classNumber); *classId = classNumber.x; } void Classfication(char* imgpath, char* result) { string res = ""; string modelTxt = "C:\\Users\\ATWER\\Desktop\\lenet_train_test.prototxt"; string modelBin = "C:\\Users\\ATWER\\Desktop\\lenet_iter_10000.caffemodel"; //string imgFileName = "C:\\Users\\ATWER\\Desktop\\9.png"; string imgFileName = imgpath; //read image Mat imgSrc = imread(imgFileName); if (imgSrc.empty()) { cout << "Failed to read image " << imgFileName << endl; exit(-1); } Mat img; cvtColor(imgSrc, img, COLOR_BGR2GRAY); //LeNet accepts 28*28 gray image resize(img, img, Size(28, 28)); bitwise_not(img, img); img /= 255; //transfer image(1*28*28) to blob data with 4 dimensions(1*1*28*28) Mat inputBlob = dnn::blobFromImage(img); dnn::Net net; try { net = dnn::readNetFromCaffe(modelTxt, modelBin); } catch (cv::Exception& ee) { cerr << "Exception: " << ee.what() << endl; if (net.empty()) { cout << "Can't load the network by using the flowing files:" << endl; cout << "modelTxt: " << modelTxt << endl; cout << "modelBin: " << modelBin << endl; exit(-1); } } Mat pred; net.setInput(inputBlob, "data");//set the network input, "data" is the name of the input layer pred = net.forward("prob");//compute output, "prob" is the name of the output layer int classId;
double classProb;
getMaxClass(pred, &classId, &classProb); res += to_string(classId); res += '|'; res += to_string(classProb); strcpy_s(result, 15, res.c_str()); }
4. 调用动态链接库
根据数据的长度申请非托管空间参考：https://blog.csdn.net/xiaoyong_net/article/details/50178021
文中说：“一定要加1,否则后面是乱码，原因未找到 ”，应该是打印字符串时会打印到“\n”为止，没有遇到\n会一直打印下去。.Length方法没有计算"\n"，+1的空间用于存放“\n”。
using System; using System.Runtime.InteropServices; namespace Test { class Program { [DllImport("E:/c++project/caffedll/x64/Debug/caffedll.dll", EntryPoint = "Classfication")] unsafe private static extern void Classfication(IntPtr imgpath, IntPtr result); private static IntPtr mallocIntptr(string strData) { //先将字符串转化成字节方式 Byte[] btData = System.Text.Encoding.Default.GetBytes(strData); //申请非拖管空间 IntPtr m_ptr = Marshal.AllocHGlobal(btData.Length); //给非拖管空间清０ Byte[] btZero = new Byte[btData.Length + 1]; //一定要加1,否则后面是乱码，原因未找到 Marshal.Copy(btZero, 0, m_ptr, btZero.Length); //给指针指向的空间赋值 Marshal.Copy(btData, 0, m_ptr, btData.Length); return m_ptr; } private static IntPtr mallocIntptr(int length) { //申请非拖管空间 IntPtr m_ptr = Marshal.AllocHGlobal(length); //给非拖管空间清０ Byte[] btZero = new Byte[length]; Marshal.Copy(btZero, 0, m_ptr, btZero.Length); //给指针指向的空间赋值 Marshal.Copy(btZero, 0, m_ptr, length); return m_ptr; } static void Main(string[] args) { string s = "C:\\Users\\ATWER\\Desktop\\9.png"; IntPtr ptrFileName; IntPtr res; //根据数据的长度申请非托管空间 ptrFileName = mallocIntptr(s); res = mallocIntptr(50); Classfication(ptrFileName, res); string result = Marshal.PtrToStringAnsi(res); string[] a = result.Split('|'); Console.WriteLine("class:"+a[0]+"\n"+"score:"+a[1]); Marshal.FreeHGlobal(res); } } }
