自建VOC数据集xml文件解析与操作实例(统计、删除、修改类别)附各实例程序代码

对xml的解析与操作在《python学习(二) ElementTree解析、读写、创建xml文件》一文中介绍完毕

本文主要是使用ElementTree对VOC数据集的xml操作的实例化

并附带一些实用小程序供大家参考使用，程序代码在下面的github网站中

https://github.com/A-mockingbird/VOCtype-datasetOperation

里面有统计VOC数据集各类别目标框数量、删除某一类别或修改某一类别名称等操作的代码

1.介绍一个简单的实例：统计自己创建voc数据集中各类别目标框数量

（1）import所需的库

import io

import sys

import os

import xml.etree.ElementTree as ET

如遇到编码问题

sys.stdout=io.TextIOWrapper(sys.stdout.buffer,encoding='utf8')

(2)解析xml文件

filepath为xml文件名

anno_path存储xml文件的文件夹路径

获取xml树结构

对xml树进行遍历，获取object(即目标)子类

对object中的目标框类别、位置信息解析，存入annos列表

返回annos及此xml文件中目标框数量num

def _ParseAnnotation(filepath):
    if os.path.exists(anno_path + filepath) == False:
        print(filepath+' :not found')
    tree = ET.parse(anno_path + filepath)
    annos = [None]*30
    num = 0 
    for annoobject in tree.iter():
        if 'object' in annoobject.tag:
            for element in list(annoobject):
                if 'name' in element.tag:
                    name = element.text 
                if 'bndbox' in element.tag:
                    for size in list(element):
                        if 'xmin' in size.tag:
                            xmin = size.text
                        if 'ymin' in size.tag:
                            ymin = size.text
                        if 'xmax' in size.tag:
                            xmax = size.text
                        if 'ymax' in size.tag:
                            ymax = size.text
                            annos[num] = {'name':name, 'xmin':int(xmin), 'ymin':int(ymin), 'xmax':int(xmax), 'ymax':int(ymax)}
                            #annos[num] = {'name':name, 'xmin':xmin, 'ymin':ymin, \
                             #           'xmax':xmax, 'ymax':ymax}                     
                            num += 1
    return num, annos

(3)统计各类别目标框数量

num为之前获取的当前xml文件的目标框数量，annos存储目标框类别、位置信息，

anno_num为存储的各类别目标框总数

class_name存储自己创建的VOC数据集类别名称

def _Count(num, annos, anno_num):
    for i in range(num):
        for j in range(len(class_name)):
            if annos[i]['name'] == class_name[j]:
                anno_num[j] += 1
    return anno_num

(4)批量删除或修改类别

import io
import sys
import os
import xml.etree.ElementTree as ET
#sys.stdout=io.TextIOWrapper(sys.stdout.buffer,encoding='utf8')
#删除或者修改数据集某一类的xml文件
##修改下面的地址为你存放xml文件的位置，注意斜杠使用/,最后末尾需要加上/
#old_annotation是修要修改的标签名，new_annotation是修改后的标签名字
anno_path = 'F:/数据集/MOCOD遮挡目标/MOCOD/Annotations-withoutocclusion/'
old_annotation = 'normal insulator'
new_annotation = 'normal single insulator'
del_annotations = ['occlusion']
#replace = True使用替换功能
#replace = False使用删除功能
REPLACE = False

def _main():
    filelist = os.listdir(anno_path)
    i = 0
    if REPLACE == True:
        for file in filelist:
            n_ = _Replace_Annotation(file)
            if n_ > 0:
                i += 1
    else:
        for file in filelist:
            n_ = _Del_Annotation(file)
            if n_ >0:
                i += 1
    print('the number of xmlfile is :' + str(i))


def _Replace_Annotation(filepath):
    if os.path.exists(anno_path + filepath) == False:
        print(filepath+' :not found')
    #建立xml树状结构
    i = 0
    while Replace_(filepath) == False:
        i += 1

    return i

def Replace_(filepath):
    if os.path.exists(anno_path + filepath) == False:
        print(filepath+' :not found')
    #建立xml树状结构
    tree = ET.parse(anno_path + filepath)
    #遍历xml文件 查找'name'
    for annoobject in tree.iter():
        if 'object' in annoobject.tag:
            for element in list(annoobject):
                if 'name' in element.tag:
                    #替换标签
                    if element.text == old_annotation:
                        element.text = new_annotation
                        print(filepath)
                        #重新写入xml，使修改生效
                        tree.write(anno_path+filepath, encoding="utf-8", xml_declaration=True)
                        return False
    return True

def _Del_Annotation(filepath):
    if os.path.exists(anno_path + filepath) == False:
        print(filepath+' :not found')
    #建立xml树状结构
    i = 0
    while Delete_(filepath) == False:
        i += 1
    return i
    
def Delete_(filepath):
    if os.path.exists(anno_path + filepath) == False:
        print(filepath+' :not found')
    #建立xml树状结构
    tree = ET.parse(anno_path + filepath)
    #遍历xml文件 查找'name'
    root = tree.getroot()
    for annoobject in root.iter():
        if 'object' in annoobject.tag:
            for element in list(annoobject):
                if 'name' in element.tag:
                    #删除标签
                    for anno in del_annotations:
                        if element.text == anno:
                            #从根节点下删除第一个子节点
                            root.remove(annoobject)
                            print(filepath)
                            #重新写入xml，使修改生效
                            tree = ET.ElementTree(root)
                            tree.write(anno_path+filepath, encoding="utf-8", xml_declaration=True)
                            return False
    return True

if __name__ == '__main__':
    _main()

github：https://github.com/A-mockingbird/VOCxml_operator/blob/master/CountObject.py