.txt(yolo)和.xml(VOC)数据集标签格式互相转换

前言：博主在处理目标检测数据集时，经常需要将标签进行不同格式的转换，因此整理了以下代码，以供参考和保存。

一、实现将txt格式转换成xml格式：

1.  
   # 将txt格式转换成xml格式数据集
2.  
   from xml.dom.minidom import Document
3.  
   import os
4.  
   import cv2
5.  
    
6.  
    
7.  
   def makexml(picPath, txtPath, xmlPath): # txt所在文件夹路径，xml文件保存路径，图片所在文件夹路径
8.  
   """此函数用于将yolo格式txt标注文件转换为voc格式xml标注文件
9.  
   在自己的标注图片文件夹下建三个子文件夹，分别命名为picture、txt、xml
10.  
    """
11.  
    #创建字典用来对类型进行转换，要与classes.txt文件中的类对应，且顺序要一致
12.  
    dic = {'0': "person", '1': "car", '2': "bike", '3': "color_cone",
13.  
    '4': "car_stop", '5': "bump", '6': "hole", '7': "animal"}
14.  
    files = os.listdir(txtPath)
15.  
    for i, name in enumerate(files):
16.  
    xmlBuilder = Document()
17.  
    annotation = xmlBuilder.createElement("annotation") # 创建annotation标签
18.  
    xmlBuilder.appendChild(annotation)
19.  
    txtFile = open(txtPath name)
20.  
    txtList = txtFile.readlines()
21.  
    img = cv2.imread(picPath name[0:-4] ".png") # 注意这里的图片后缀，.jpg/.png
22.  
    Pheight, Pwidth, Pdepth = img.shape
23.  
     
24.  
    folder = xmlBuilder.createElement("folder") # folder标签
25.  
    foldercontent = xmlBuilder.createTextNode("datasetRGB")
26.  
    folder.appendChild(foldercontent)
27.  
    annotation.appendChild(folder)
28.  
     
29.  
    filename = xmlBuilder.createElement("filename") # filename标签
30.  
    filenamecontent = xmlBuilder.createTextNode(name[0:-4] ".png")
31.  
    filename.appendChild(filenamecontent)
32.  
    annotation.appendChild(filename)
33.  
     
34.  
    size = xmlBuilder.createElement("size") # size标签
35.  
    width = xmlBuilder.createElement("width") # size子标签width
36.  
    widthcontent = xmlBuilder.createTextNode(str(Pwidth))
37.  
    width.appendChild(widthcontent)
38.  
    size.appendChild(width)
39.  
     
40.  
    height = xmlBuilder.createElement("height") # size子标签height
41.  
    heightcontent = xmlBuilder.createTextNode(str(Pheight))
42.  
    height.appendChild(heightcontent)
43.  
    size.appendChild(height)
44.  
     
45.  
    depth = xmlBuilder.createElement("depth") # size子标签depth
46.  
    depthcontent = xmlBuilder.createTextNode(str(Pdepth))
47.  
    depth.appendChild(depthcontent)
48.  
    size.appendChild(depth)
49.  
     
50.  
    annotation.appendChild(size)
51.  
     
52.  
    for j in txtList:
53.  
    oneline = j.strip().split(" ")
54.  
    object = xmlBuilder.createElement("object") # object 标签
55.  
    picname = xmlBuilder.createElement("name") # name标签
56.  
    namecontent = xmlBuilder.createTextNode(dic[oneline[0]])
57.  
    picname.appendChild(namecontent)
58.  
    object.appendChild(picname)
59.  
     
60.  
    pose = xmlBuilder.createElement("pose") # pose标签
61.  
    posecontent = xmlBuilder.createTextNode("Unspecified")
62.  
    pose.appendChild(posecontent)
63.  
    object.appendChild(pose)
64.  
     
65.  
    truncated = xmlBuilder.createElement("truncated") # truncated标签
66.  
    truncatedContent = xmlBuilder.createTextNode("0")
67.  
    truncated.appendChild(truncatedContent)
68.  
    object.appendChild(truncated)
69.  
     
70.  
    difficult = xmlBuilder.createElement("difficult") # difficult标签
71.  
    difficultcontent = xmlBuilder.createTextNode("0")
72.  
    difficult.appendChild(difficultcontent)
73.  
    object.appendChild(difficult)
74.  
     
75.  
    bndbox = xmlBuilder.createElement("bndbox") # bndbox标签
76.  
    xmin = xmlBuilder.createElement("xmin") # xmin标签
77.  
    mathData = int(((float(oneline[1])) * Pwidth 1) - (float(oneline[3])) * 0.5 * Pwidth)
78.  
    xminContent = xmlBuilder.createTextNode(str(mathData))
79.  
    xmin.appendChild(xminContent)
80.  
    bndbox.appendChild(xmin)
81.  
     
82.  
    ymin = xmlBuilder.createElement("ymin") # ymin标签
83.  
    mathData = int(((float(oneline[2])) * Pheight 1) - (float(oneline[4])) * 0.5 * Pheight)
84.  
    yminContent = xmlBuilder.createTextNode(str(mathData))
85.  
    ymin.appendChild(yminContent)
86.  
    bndbox.appendChild(ymin)
87.  
     
88.  
    xmax = xmlBuilder.createElement("xmax") # xmax标签
89.  
    mathData = int(((float(oneline[1])) * Pwidth 1) (float(oneline[3])) * 0.5 * Pwidth)
90.  
    xmaxContent = xmlBuilder.createTextNode(str(mathData))
91.  
    xmax.appendChild(xmaxContent)
92.  
    bndbox.appendChild(xmax)
93.  
     
94.  
    ymax = xmlBuilder.createElement("ymax") # ymax标签
95.  
    mathData = int(((float(oneline[2])) * Pheight 1) (float(oneline[4])) * 0.5 * Pheight)
96.  
    ymaxContent = xmlBuilder.createTextNode(str(mathData))
97.  
    ymax.appendChild(ymaxContent)
98.  
    bndbox.appendChild(ymax)
99.  
     
100.  
     object.appendChild(bndbox) # bndbox标签结束
101.  
      
102.  
     annotation.appendChild(object)
103.  
      
104.  
     f = open(xmlPath name[0:-4] ".xml", 'w')
105.  
     xmlBuilder.writexml(f, indent='\t', newl='\n', addindent='\t', encoding='utf-8')
106.  
     f.close()
107.  
      
108.  
      
109.  
     if __name__ == "__main__":
110.  
     picPath = "C:\\data\\ir_det_dataset\\Images\\fir\\" # 图片所在文件夹路径，后面的\\一定要带上
111.  
     txtPath = "C:\\data\\ir_det_dataset\\labels\\fir\\" # txt所在文件夹路径，后面的\\一定要带上
112.  
     xmlPath = "C:\\data\\ir_det_dataset\\xml\\fir\\" # xml文件保存路径，后面的\\一定要带上
113.  
     makexml(picPath, txtPath, xmlPath)

转换之后第一行可能会出现<?xml version="1.0" encoding="utf-8"?>，这行代码声明所用的xml版本是1.0，用xml传输数据时的字符编码是utf-8，如果想去除第一行可以采用以下代码：

1.  
   import os
2.  
   def file_name(input_dir,output_dir,spile):
3.  
   for root, dirs, files in os.walk(input_dir):
4.  
   for item in files:
5.  
   f = open(input_dir item, "r", encoding='UTF-8')
6.  
   content = f.read()
7.  
   content = content.replace('<?xml version="1.0" encoding="utf-8"?>', spile)
8.  
   with open(os.path.join(output_dir, item), 'w',encoding='UTF-8') as fval:
9.  
   fval.write(content)
10.  
    f.close()
11.  
    if __name__ == '__main__':
12.  
    input_dir = "C:\\data\\ir_det_dataset\\xml\\fir\\" #旧文件夹
13.  
    output_dir = "C:\\data\\ir_det_dataset\\xml\\fir\\xml\\" #新文件夹
14.  
    file_name(input_dir, output_dir, "")

二、实现将txt格式转换成xml格式：

1.  
   import xml.etree.ElementTree as ET
2.  
   import os
3.  
    
4.  
   classes = ["person", "car", "bike", "color_cone", "car_stop", "bump", "hole", "animal"]
5.  
    
6.  
   # 和自己classes.txt中的类别要一个一个地对应
7.  
    
8.  
    
9.  
   def convert(size, box):
10.  
    dw = 1. / size[0]
11.  
    dh = 1. / size[1]
12.  
    x = (box[0] box[1]) / 2.0
13.  
    y = (box[2] box[3]) / 2.0
14.  
    w = box[1] - box[0]
15.  
    h = box[3] - box[2]
16.  
    x = x * dw
17.  
    w = w * dw
18.  
    y = y * dh
19.  
    h = h * dh
20.  
    return (x, y, w, h)
21.  
     
22.  
     
23.  
    def convert_annotation(image_id):
24.  
     
25.  
    in_file = open("C:\\data\\ir_det_dataset\\Images\\fir\\xml\\%s.xml" % (image_id), encoding='UTF-8')
26.  
    # in_file输入要转换的文件路径
27.  
     
28.  
    out_file = open("C:\\data\\ir_det_dataset\\Images\\fir\\txt\\%s.txt" % (image_id), 'w')
29.  
    # out_file输出转换后的文件路径
30.  
     
31.  
    tree = ET.parse(in_file)
32.  
    root = tree.getroot()
33.  
    size = root.find('size')
34.  
    w = int(size.find('width').text)
35.  
    h = int(size.find('height').text)
36.  
     
37.  
    for obj in root.iter('object'):
38.  
    cls = obj.find('name').text
39.  
    # print(cls)
40.  
    if cls not in classes:
41.  
    continue
42.  
    cls_id = classes.index(cls)
43.  
    xmlbox = obj.find('bndbox')
44.  
    b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text), float(xmlbox.find('ymin').text),
45.  
    float(xmlbox.find('ymax').text))
46.  
    bb = convert((w, h), b)
47.  
    out_file.write(str(cls_id) " " " ".join([str(a) for a in bb]) '\n')
48.  
     
49.  
     
50.  
    CURRENT_DIR = os.path.dirname(os.path.abspath(__file__))
51.  
    xml_path = os.path.join(CURRENT_DIR, "C:\\data\\ir_det_dataset\\Images\\fir\\xml\\")
52.  
    # xml文件路径
53.  
     
54.  
    # xml list
55.  
    img_xmls = os.listdir(xml_path)
56.  
    for img_xml in img_xmls:
57.  
    label_name = img_xml.split('.')[0]
58.  
    print(label_name)
59.  
    convert_annotation(label_name)

以上代码同时参考了其他博主，第一次发帖如有侵权请联系删除~

这篇好文章是转载于：学新通技术网