【caffe】OpenCV Load caffe model（转）

原文地址：http://blog.csdn.net/guduruyu/article/details/76006003

 

上一篇，我们介绍了opencv_contrib中的模块在windows下的编译，也提到了其中的dnn模块可以读取caffe的训练模型用于目标检测，这里我们具体介绍一下如何使用dnn读取caffe模型并进行目标分类。

 

代码如下:(代码主要来自参考[2]和[3]):

 

1. #include <opencv2/dnn.hpp>

2. #include <opencv2/imgproc.hpp>

3. #include <opencv2/highgui.hpp>

4.  

5. #include <fstream>

6. #include <iostream>

7. #include <cstdlib>

8.  

9. /* Find best class for the blob (i. e. class with maximal probability) */

10. void getMaxClass(cv::dnn::Blob &probBlob, int *classId, double *classProb)

11. {

12. cv::Mat probMat = probBlob.matRefConst().reshape(1, 1); //reshape the blob to 1x1000 matrix

13. cv::Point classNumber;

14.  

15. cv::minMaxLoc(probMat, NULL, classProb, NULL, &classNumber);

16. *classId = classNumber.x;

17. }

18.  

19. std::vector<cv::String> readClassNames(const char *filename = "synset_words.txt")

20. {

21. std::vector<cv::String> classNames;

22.  

23. std::ifstream fp(filename);

24. if (!fp.is_open())

25. {

26. std::cerr << "File with classes labels not found: " << filename << std::endl;

27. exit(-1);

28. }

29.  

30. std::string name;

31. while (!fp.eof())

32. {

33. std::getline(fp, name);

34. if (name.length())

35. classNames.push_back(name.substr(name.find(' ') + 1));

36. }

37.  

38. fp.close();

39. return classNames;

40. }

41.  

42. int main(int argc, char **argv)

43. {

44. void cv::dnn::initModule();

45.  

46. cv::String modelTxt = "bvlc_googlenet.prototxt";

47. cv::String modelBin = "bvlc_googlenet.caffemodel";

48. cv::String imageFile = "space_shuttle.jpg";

49.  

50. cv::dnn::Net net = cv::dnn::readNetFromCaffe(modelTxt, modelBin);

51.  

52. if (net.empty())

53. {

54. std::cerr << "Can't load network by using the following files: " << std::endl;

55. std::cerr << "prototxt: " << modelTxt << std::endl;

56. std::cerr << "caffemodel: " << modelBin << std::endl;

57. std::cerr << "bvlc_googlenet.caffemodel can be downloaded here:" << std::endl;

58. std::cerr << "http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel" << std::endl;

59. exit(-1);

60. }

61.  

62. //! [Prepare blob]

63. cv::Mat img = cv::imread(imageFile, cv::IMREAD_COLOR);

64. if (img.empty())

65. {

66. std::cerr << "Can't read image from the file: " << imageFile << std::endl;

67. exit(-1);

68. }

69.  

70. cv::resize(img, img, cv::Size(224, 224));

71. cv::dnn::Blob inputBlob = cv::dnn::Blob(img); //Convert Mat to dnn::Blob image batch

72. //! [Prepare blob]

73.  

74. //! [Set input blob]

75. net.setBlob(".data", inputBlob); //set the network input

76. //! [Set input blob]

77.  

78. //! [Make forward pass]

79. net.forward(); //compute output

80. //! [Make forward pass]

81.  

82. //! [Gather output]

83. cv::dnn::Blob prob = net.getBlob("prob"); //gather output of "prob" layer

84.  

85. int classId;

86. double classProb;

87. getMaxClass(prob, &classId, &classProb);//find the best class

88. //! [Gather output]

89.  

90. //! [Print results]

91. std::vector<cv::String> classNames = readClassNames();

92. std::cout << "Best class: #" << classId << " '" << classNames.at(classId) << "'" << std::endl;

93. std::cout << "Probability: " << classProb * 100 << "%" << std::endl;

94.  

95. //! [Print results]

96.  

97. return 0;

98. } //main

 

 

代码详解:
1、首先需要下载GoogLeNet模型及分类相关文件，可以从官网下载（或复制粘贴）：  bvlc_googlenet.prototxt、bvlc_googlenet.caffemodel以及synset_words.txt.也可以直接下载我长传的打包好的资源（包括了2中的图片）

2、下载待检测图片文件，如下：

Buran space shuttle

3、读取.protxt文件和.caffemodel文件：

 

cv::dnn::Net net = cv::dnn::readNetFromCaffe(modelTxt, modelBin);

 

 

 

4、检查网络是否读取成功：

 

1. if (net.empty())

2. {

3. std::cerr << "Can't load network by using the following files: " << std::endl;

4. std::cerr << "prototxt: " << modelTxt << std::endl;

5. std::cerr << "caffemodel: " << modelBin << std::endl;

6. std::cerr << "bvlc_googlenet.caffemodel can be downloaded here:" << std::endl;

7. std::cerr << "http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel" << std::endl;

8. exit(-1);

9. }

5、读取图片并将其转换成GoogleNet可以读取的blob：

 

 

1. cv::Mat img = cv::imread(imageFile, cv::IMREAD_COLOR);

2. if (img.empty())

3. {

4. std::cerr << "Can't read image from the file: " << imageFile << std::endl;

5. exit(-1);

6. }

7.  

8. cv::resize(img, img, cv::Size(224, 224));

9. cv::dnn::Blob inputBlob = cv::dnn::Blob(img); //Convert Mat to dnn::Blob image batch

 

 

 

 

 

6、将blob传递给网络：

 

	net.setBlob(".data", inputBlob);        //set the network input  

 

 

 

7、前向传递：

 

	net.forward();                          //compute output  

 

 

 

8、分类：

 

	getMaxClass(prob, &classId, &classProb);//find the best class  

 

 

 

9、打印分类结果：

 

1. std::vector<cv::String> classNames = readClassNames();

2. std::cout << "Best class: #" << classId << " '" << classNames.at(classId) << "'" << std::endl;

3. std::cout << "Probability: " << classProb * 100 << "%" << std::endl;

 

 

 

运行,报错如下:

 

找了很久，终于在参考[3]中找到了解决方案，原因是这里将图像数据转换成blob的方法来自于老版本，在新版本中不兼容。解决方法如下：将cv::dnn::Blob(img) 用cv::dnn::Blob::fromImages(img)替换掉。

 

修改后，再运行，结果如下：

 

参考：

[1] http://docs.opencv.org/trunk/d5/de7/tutorial_dnn_googlenet.html
[2] http://blog.csdn.net/langb2014/article/details/50555910
[3] https://github.com/opencv/opencv_contrib/issues/749

 

-----------------------------------------

2017.07.24