devbox
从前慢现在也慢
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

使用自己训练的yolov3或yolov4模型自动标注成voc格式数据_yolov4标注格式

作者：从前慢现在也慢 | 2024-06-16 20:03:41

赞

踩

yolov4标注格式

yolo数据格式转voc格式：

  1. #!/usr/bin/env python
  2. # -*- coding: utf-8 -*-
  3. # @Time        :2021/8/31 下午3:06
  4. # @Author      :
  5. # @ProjectName :mmdetection-master
  6. # @File        :yolo2voc.py
  7. # @Description :
  8. import os
  9. import codecs
  10. import cv2
  11.  
  12. def vocXMLFormat(filePath, imgName, imgW, imgH, objNames, locs, depth=3, truncated=0, difficult=0):
  13.     """
  14.     Generate xml annotation file,eg:
  15.         vocXMLFormat("0000.xml", "0001.png", 608, 608, ["person", "TV"], [[24, 32, 156, 145], [124, 76, 472, 384]])
  16.     Args:
  17.         filePath:The path of the file saved in the generated xml
  18.         imgName:xml file annotates the name of the picture
  19.         imgW:
  20.         imgH:
  21.         objNames:The object contained in the picture, format: ["object label 1", "object label 2"...]
  22.         locs:The picture contains the coordinates of the object, format: [[x,y,w,h],[x,y,w,h]...]
  23.         depth:
  24.         truncated:
  25.         difficult:
  27.     Returns:
  29.     """
  30.     if (objNames == None) or (locs == None):
  31.         print("The objNames or locs is None!!!")
  32.         return
  33.     with codecs.open(filePath, 'w', 'utf-8') as xml:
  34.         xml.write('<?xml version="1.0" encoding="UTF-8"?>\n')
  35.         xml.write('<annotation>\n')
  36.         xml.write('\t<folder>' + 'voc format' + '</folder>\n')
  37.         xml.write('\t<filename>' + imgName + '</filename>\n')
  38.         xml.write('\t<path>' + imgName + '</path>\n')
  39.         xml.write('\t<source>\n')
  40.         xml.write('\t\t<database>weiz</database>\n')
  41.         xml.write('\t</source>\n')
  42.         xml.write('\t<size>\n')
  43.         xml.write('\t\t<width>' + str(imgW) + '</width>\n')
  44.         xml.write('\t\t<height>' + str(imgH) + '</height>\n')
  45.         xml.write('\t\t<depth>' + str(depth) + '</depth>\n')
  46.         xml.write('\t</size>\n')
  47.         xml.write('\t<segmented>0</segmented>\n')
  48.  
  49.         for ind, name in enumerate(objNames):
  50.             xml.write('\t<object>\n')
  51.             xml.write('\t\t<name>' + name + '</name>\n')
  52.             xml.write('\t\t<pose>Unspecified</pose>\n')
  53.             xml.write('\t\t<truncated>' + str(truncated) + '</truncated>\n')
  54.             xml.write('\t\t<difficult>' + str(difficult) + '</difficult>\n')
  55.             xml.write('\t\t<bndbox>\n')
  56.             xml.write('\t\t\t<xmin>' + str(locs[ind][0]) + '</xmin>\n')
  57.             xml.write('\t\t\t<ymin>' + str(locs[ind][1]) + '</ymin>\n')
  58.             xml.write('\t\t\t<xmax>' + str(locs[ind][0] + locs[ind][2]) + '</xmax>\n')
  59.             xml.write('\t\t\t<ymax>' + str(locs[ind][1] + locs[ind][3]) + '</ymax>\n')
  60.             xml.write('\t\t</bndbox>\n')
  61.             xml.write('\t</object>\n')
  62.  
  63.         xml.write('</annotation>')
  64.     xml.close()
  65.     print("The {} accomplish!".format(filePath))
  66.  
  67.  
  68. def read_line(file_name):
  69.     """
  70.     read file by line
  71.     :param file_name:
  72.     :return:
  73.     """
  74.     lines = []
  75.     file = open(file_name)
  76.     for line in file:
  77.         line = line.strip('\n')
  78.         lines.append(line)
  79.     file.close()
  80.     return lines
  81.  
  82.  
  83. def showAnnotions(image, locs, labels):
  84.     """
  85.     show annotions
  86.     :param locs:
  87.     :param labels:
  88.     :return:
  89.     """
  90.     for ind, (x, y, w, h) in enumerate(locs):
  91.         cv2.rectangle(image, (x, y), (x + w, y + h), (0, 0, 255), 2)
  92.         cv2.putText(image, labels[ind], (x, y), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 1)
  93.  
  94.     cv2.imshow("iamge", image)
  95.     cv2.waitKey(30)
  96.  
  97.  
  98. txt_path = "/home/zyy/桌面/label"
  99. image_path = "/home/zyy/桌面/images"
  100. save_xml_path = "/home/zyy/桌面/xml"
  101. def main():
  102.     txt_list = os.listdir(txt_path)
  103.     label_list = read_line("/home/zyy/桌面/label.txt")
  104.     for txt_name in txt_list:
  105.         image_name, _ = os.path.splitext(txt_name)
  106.         for extension in ["png", "jpg"]:  # read image
  107.             image_name_path = image_name + '.' + extension
  108.             image_name_path = os.path.join(image_path, image_name_path)
  109.             if os.path.exists(image_name_path):
  110.                 image = cv2.imread(image_name_path)
  111.                 image_h, image_w, image_channel = image.shape
  112.                 break
  113.             else:
  114.                 continue
  115.  
  116.         txt_name_path = os.path.join(txt_path, txt_name)
  117.         lines = read_line(txt_name_path)
  118.         object_locs = []
  119.         object_labels = []
  120.         for line in lines:
  121.             ind, center_x, center_y, w, h = line.split(' ')
  122.             x = (float(center_x) - float(w) / 2) * image_w
  123.             y = (float(center_y) - float(h) / 2) * image_h
  124.             w = float(w) * image_w
  125.             h = float(h) * image_h
  126.             object_locs.append([int(x), int(y), int(w), int(h)])
  127.             object_labels.append(label_list[int(ind)])
  128.  
  129.         # showAnnotions(image, object_locs, object_labels)
  130.         xml_name_path = os.path.join(save_xml_path, image_name + ".xml")
  131.         vocXMLFormat(xml_name_path, os.path.split(image_name_path)[-1], image_w, image_h, object_labels, object_locs, image_channel)
  132.  
  133.  
  134. if __name__ == "__main__":
  135.     main()

信息都在代码里，只需要自己实现getLabelInfo(img)检测部分即可，也可以参考我写的：

  1. #!/usr/bin/env python
  2. # -*- coding: utf-8 -*-
  3. # @Time        :2020/11/3 14:00
  4. # @Author      :weiz
  5. # @ProjectName :autoLabeling
  6. # @File        :dataLabeling.py
  7. # @Description :数据标注文件
  8. import os
  9. import codecs
  10. import cv2
  11.  
  12. import yoloMain
  13.  
  14.  
  15. def vocXMLFormat(filePath, imgName, imgW, imgH, objNames, locs, depth=3, truncated=0, difficult=0):
  16.     """
  17.     生成xml数据文件;eg:
  18.         vocXMLFormat("0000.xml", "0001.png", 608, 608, ["person", "TV"], [[24, 32, 156, 145], [124, 76, 472, 384]])
  19.     :param filePath: 生成xml所保存文件的路径
  20.     :param imgName: xml文件标注图片的名字
  21.     :param imgW: 图片的宽
  22.     :param imgH: 图片的高
  23.     :param objNames: 图片包含的目标，格式：["目标1","目标2"...]
  24.     :param locs: 图片包含目标的坐标，格式：[[x,y,w,h],[x,y,w,h]...]
  25.     :param depth: 图片的深度，默认是3
  26.     :param truncated: 是否被截断（0表示完整）
  27.     :param difficult: 目标是否难以识别（0表示容易识别）
  28.     :return:
  29.     """
  30.     if (objNames == None) or (locs == None):
  31.         print("The objNames or locs is None!!!")
  32.         return
  33.     with codecs.open(filePath, 'w', 'utf-8') as xml:
  34.         xml.write('<?xml version="1.0" encoding="UTF-8"?>\n')
  35.         xml.write('<annotation>\n')
  36.         xml.write('\t<folder>' + 'voc format' + '</folder>\n')
  37.         xml.write('\t<filename>' + imgName + '</filename>\n')
  38.         xml.write('\t<path>' + imgName + '</path>\n')
  39.         xml.write('\t<source>\n')
  40.         xml.write('\t\t<database>weiz</database>\n')
  41.         xml.write('\t</source>\n')
  42.         xml.write('\t<size>\n')
  43.         xml.write('\t\t<width>' + str(imgW) + '</width>\n')
  44.         xml.write('\t\t<height>' + str(imgH) + '</height>\n')
  45.         xml.write('\t\t<depth>' + str(depth) + '</depth>\n')
  46.         xml.write('\t</size>\n')
  47.         xml.write('\t<segmented>0</segmented>\n')
  48.  
  49.         for ind, name in enumerate(objNames):
  50.             xml.write('\t<object>\n')
  51.             xml.write('\t\t<name>' + name + '</name>\n')
  52.             xml.write('\t\t<pose>Unspecified</pose>\n')
  53.             xml.write('\t\t<truncated>' + str(truncated) + '</truncated>\n')
  54.             xml.write('\t\t<difficult>' + str(difficult) + '</difficult>\n')
  55.             xml.write('\t\t<bndbox>\n')
  56.             xml.write('\t\t\t<xmin>' + str(locs[ind][0]) + '</xmin>\n')
  57.             xml.write('\t\t\t<ymin>' + str(locs[ind][1]) + '</ymin>\n')
  58.             xml.write('\t\t\t<xmax>' + str(locs[ind][0] + locs[ind][2]) + '</xmax>\n')
  59.             xml.write('\t\t\t<ymax>' + str(locs[ind][1] + locs[ind][3]) + '</ymax>\n')
  60.             xml.write('\t\t</bndbox>\n')
  61.             xml.write('\t</object>\n')
  62.  
  63.         xml.write('</annotation>')
  64.     xml.close()
  65.     print("The {} accomplish!".format(filePath))
  66.  
  67.  
  68. def autoLabeling(yoloModel, modelSign, savePath, img, imgName=None, extraName=None, depth=3, truncated=0, difficult=0):
  69.     """
  70.     生成标注图片img的xml文件
  71.     :param yoloModel: yolov3或者yolov4
  72.     :param modelSign: 模型标识位，3表示yolov3,4表示yolov4
  73.     :param savePath: 保存路径
  74.     :param img: 图片
  75.     :param imgName: 图片的名字
  76.     :param extraName: 附加额外的名字：默认是从000001.png开始命名图片；如果该参数为hsh，则命名从hsh_000001.png开始
  77.     :param depth: 图片的深度
  78.     :param truncated: 是否被截断（0表示完整）
  79.     :param difficult: 目标是否难以识别（0表示容易识别）
  80.     :return:
  81.     """
  82.     global savePathFolderNum
  83.     if savePathFolderNum == -1:
  84.         try:
  85.             savePathFolderNum = 0
  86.             fileList = os.listdir(savePath)
  87.             for file in fileList:        # 统计png文件的个数
  88.                 if "png" in os.path.splitext(file)[-1]:
  89.                     savePathFolderNum = savePathFolderNum + 1
  90.         except FileNotFoundError:
  91.             os.mkdir(savePath)
  92.             savePathFolderNum = 0
  93.     if imgName == None:                  # 确定保存图片的名字
  94.         if extraName != None:
  95.             imgName = extraName + '_' + "{:0>6d}".format(savePathFolderNum)
  96.         else:
  97.             imgName = "{:0>6d}".format(savePathFolderNum)
  98.     else:
  99.         if extraName != None:
  100.             imgName = extraName + '_' + imgName
  101.  
  102.     imgPath = os.path.join(savePath, imgName + ".png")
  103.     xmlPath = os.path.join(savePath, imgName + ".xml")
  104.  
  105.     locs, labels = getLabelInfo(yoloModel, modelSign, img)
  106.     if len(labels) > 0:
  107.         savePathFolderNum = savePathFolderNum + 1
  108.         vocXMLFormat(xmlPath, imgName + ".png", img.shape[1], img.shape[0], labels, locs, depth, truncated, difficult)
  109.         cv2.imwrite(imgPath, img)
  110.  
  111.     return locs, labels
  112.  
  113.  
  114. def getLabelInfo(yoloModel, modelSign, img):
  115.     """
  116.     实现图片的检测，并返回标注信息,eg:
  117.         labels = ["person", "TV"]
  118.         locs = [[24, 32, 156, 145], [124, 76, 472, 384]]
  119.     :param yoloModel:
  120.     :param modelSign:
  121.     :param img:
  122.     :return: 返回labels和locs，格式：["目标1","目标2"...]和[[x,y,w,h],[x,y,w,h]...]
  123.     """
  124.     if modelSign == 3:
  125.         boxes, labels, confs, timeLabel = yoloMain.runningYolov3(yoloModel, img)
  126.     else:
  127.         boxes, labels, confs, timeLabel = yoloMain.runningYolov4(yoloModel, img)
  128.     locs = []
  129.     if boxes == []:
  130.         return locs, labels
  131.     for x1, y1, x2, y2 in boxes:
  132.         w = x2 - x1
  133.         h = y2 - y1
  134.         locs.append([x1, y1, w, h])
  135.  
  136.     return locs, labels
  137.  
  138.  
  139. def showAnnotions(image, locs, labels):
  140.     """
  141.     显示标注
  142.     :param locs:
  143.     :param labels:
  144.     :return:
  145.     """
  146.     for ind, (x, y, w, h) in enumerate(locs):
  147.         cv2.rectangle(image, (x, y), (x + w, y + h), (0, 0, 255), 2)
  148.         cv2.putText(image, labels[ind], (x, y), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 1)
  149.  
  150.  
  151. def videoAnnotation(videoPath, savePath, gap=10, yoloModels=3):
  152.     """
  153.     自动标注视频数据
  154.     :param videoPath: 视频的路径
  155.     :param savePath: 标注后保存的路径
  156.     :param gap: 每多少帧才标注
  157.     :param yoloModels: 使用yolov3或者yolov4模型，默认使用yolov3
  158.     :return:
  159.     """
  160.     if yoloModels != 4:   # 使用yolov3
  161.         yoloModels = yoloMain.getYolov3()
  162.         modelSign = 3
  163.     else:                 # 使用yolov4
  164.         yoloModels = yoloMain.getYolov4()
  165.         modelSign = 4
  166.     cap = cv2.VideoCapture(videoPath)
  167.     frameNum = 0
  168.     videoName = os.path.splitext(os.path.basename(videPath))[0]
  169.     while True:
  170.         ok, img = cap.read()
  171.         frameNum = frameNum + 1
  172.         if not ok:
  173.             break
  174.         if frameNum % gap != 0:
  175.             continue
  176.  
  177.         locs, labels = autoLabeling(yoloModels, modelSign, savePath, img, extraName=videoName)
  178.         showAnnotions(img, locs, labels)
  179.  
  180.         cv2.imshow('video', img)
  181.         if cv2.waitKey(1) & 0xFF == 27:
  182.             cap.release()  # 关闭摄像头
  183.             break
  184.  
  185.     cv2.destroyAllWindows()
  186.  
  187.  
  188. def imagesAnnotation(imagesPath, savePath, yoloModels=3):
  189.     """
  190.     图片自动标注
  191.     :param imagesPath:
  192.     :param savePath:
  193.     :param yoloModels:
  194.     :return:
  195.     """
  196.     if yoloModels != 4:   # 使用yolov3
  197.         yoloModels = yoloMain.getYolov3()
  198.         modelSign = 3
  199.     else:                 # 使用yolov4
  200.         yoloModels = yoloMain.getYolov4()
  201.         modelSign = 4
  202.  
  203.     imagesList = os.listdir(imagesPath)
  204.     for imageName in imagesList:
  205.         imagePath = os.path.join(imagesPath, imageName)
  206.         image = cv2.imread(imagePath)
  207.  
  208.         locs, labels = autoLabeling(yoloModels, modelSign, savePath, image, os.path.splitext(imageName)[0])
  209.         showAnnotions(image, locs, labels)
  210.  
  211.         cv2.imshow("image", image)
  212.         if cv2.waitKey(1) & 0xFF == 27:
  213.             break
  214.  
  215.     cv2.destroyAllWindows()
  216.  
  217.  
  218. videPath = "./videos/004.avi"
  219. imagesPath = "./srcImages"            # 待标注的图片
  220. savePath = "C:/Users/weiz/Desktop/annotions"        # 该文件可以不存在，会自动创建
  221. savePathFolderNum = -1                              # 所存路径文件的个数，-1表示还没有读取
  222. yoloMoels = 3                                       # 3表示使用yolov3模型标注，4表示使用yolov4标注
  223. if __name__ == "__main__":
  224.     #videoAnnotation(videPath, savePath, 10, yoloMoels)
  225.  
  226.     imagesAnnotation(imagesPath, savePath, yoloMoels)

        c++版本写入xml文件。

  1. #include <iostream>
  2. #include <fstream>
  3. #include <vector>
  4. #include <string>
  5.  
  6. /*
  7. imgPath:图片存储路径，eg：./abcdeg/ssf.png
  8. objNames:目标的labels
  9. locs:与objNames对应的位置信息，存储格式是[x,y,w,h]
  10. imgW:图片宽
  11. imgH:图片高
  12. */
  13. int vocXMLFormat(const std::string& imgPath, const std::vector<std::string>& objNames, const std::vector<std::vector<int>>& locs, int imgW, int imgH, int depth = 3, int truncated = 0, int difficult = 0) {
  14. 	std::string xmlPath;
  15. 	std::string imgName;
  16. 	int point_index = -1;
  17. 	int img_name_index = imgPath.length() - 1;
  18. 	while (('/' != imgPath[img_name_index]) && (img_name_index >= 0)) {
  19. 		if ((point_index < 0) && ('.' == imgPath[img_name_index])) {
  20. 			point_index = img_name_index;
  21. 		}
  22.  
  23.         // 适配windows的路径
  24. 		if ('\\' == imgPath[img_name_index]) {
  25. 			break;
  26. 		}
  27.         
  28. 		img_name_index = img_name_index - 1;
  29. 	}
  30. 	xmlPath = imgPath.substr(0, point_index) + ".xml";
  31. 	imgName = imgPath.substr(img_name_index + 1, imgPath.length());
  32. 	std::cout << "xml save path:" << xmlPath << "  img name:" << imgName << std::endl;
  33.  
  34. 	std::ofstream xmlFile(xmlPath);
  35. 	if (xmlFile) {
  36. 		std::string data = "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n";
  37. 		xmlFile << "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n";
  38. 		xmlFile << "<annotation>\n";
  39. 		xmlFile << "\t<folder>voc format</folder>\n";
  40. 		xmlFile << "\t<filename>" + imgName + "</filename>\n";
  41. 		xmlFile << "\t<path>" + imgPath + "</path>\n";
  42. 		xmlFile << "\t<source>\n";
  43. 		xmlFile << "\t\t<database>weiz</database>\n";
  44. 		xmlFile << "\t</source>\n";
  45. 		xmlFile << "\t<size>\n";
  46. 		xmlFile << "\t\t<width>" + std::to_string(imgW) + "</width>\n";
  47. 		xmlFile << "\t\t<height>" + std::to_string(imgH) + "</height>\n";
  48. 		xmlFile << "\t\t<depth>" + std::to_string(depth) + "</depth>\n";
  49. 		xmlFile << "\t</size>\n";
  50. 		xmlFile << "\t<segmented>0</segmented>\n";
  51.  
  52. 		for (int i = 0; i < objNames.size(); ++i) {
  53. 			xmlFile << "\t<object>\n";
  54. 			xmlFile << "\t\t<name>" + objNames[i] + "</name>\n";
  55. 			xmlFile << "\t\t<pose>Unspecified</pose>\n";
  56. 			xmlFile << "\t\t<truncated>" + std::to_string(truncated) + "</truncated>\n";
  57. 			xmlFile << "\t\t<difficult>" + std::to_string(difficult) + "</difficult>\n";
  58. 			xmlFile << "\t\t<bndbox>\n";
  59. 			xmlFile << "\t\t\t<xmin>" + std::to_string(locs[i][0]) + "</xmin>\n";
  60. 			xmlFile << "\t\t\t<ymin>" + std::to_string(locs[i][1]) + "</ymin>\n";
  61. 			xmlFile << "\t\t\t<xmax>" + std::to_string(locs[i][0] + locs[i][2]) + "</xmax>\n";
  62. 			xmlFile << "\t\t\t<ymax>" + std::to_string(locs[i][1] + locs[i][3]) + "</ymax>\n";
  63. 			xmlFile << "\t\t</bndbox>\n";
  64. 			xmlFile << "\t</object>\n";
  65. 		}
  66.  
  67. 		xmlFile << "</annotation>";
  68. 		xmlFile.close();
  69. 	}
  70. 	else {
  71. 		std::cerr << "无法打开文件！" << std::endl;
  72. 		return 1;
  73. 	}
  74.  
  75. 	return 0;
  76. }
  77.  
  78. int main()
  79. {
  80. 	//seg_roi();
  81.  
  82. 	std::string imgPath = "./sdfdsfe.jpeg";
  83. 	std::vector<std::string> objNames = {"sss","bbb"};
  84. 	std::vector<std::vector<int>> locs;
  85. 	std::vector<int> aa = { 1,2,3,4 };
  86. 	std::vector<int> bb = { 5,6,7,8 };
  87. 	locs.push_back(aa); locs.push_back(bb);
  88.  
  89. 	vocXMLFormat(imgPath, objNames, locs, 1920, 1080);
  90. }

声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/从前慢现在也慢/article/detail/728058
推荐阅读
相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。

  

闽ICP备14008679号