Python批量修改labelImg生成的XML文件：过滤类别与统一小写

Ada

发布时间：2026-10-13 | 5人浏览过

原创

本文介绍如何使用Python脚本批量处理labelImg生成的XML标注文件。通过解析ElementTree，程序可自动过滤掉不需要的目标类别（如仅保留people和cat），并强制将所有保留标签的名称统一转换为小写，以适配py-faster-rcnn等对标签大小写敏感的训练框架。该方法避免了重新打标的繁琐工作，高效完成数据预处理。

概述与背景

概述

在使用labelImg完成图像标注后，有时我们需要仅利用部分类别进行模型训练，而不希望重新进行全量标注并生成新的XML文件。为此，可以直接通过脚本处理现有的XML文件，删除其中不需要的标签类别及其相关属性。

在标注过程中，由于人工操作误差，标签名称可能出现大小写不一致的情况。若后续训练框架（如py-faster-rcnn）要求标签必须全部为小写，则需要在程序中增加将标签值统一转换为小写的逻辑。

以下以一个示例XML文件为基础，其中故意混入了大写标签以演示处理效果。


<annotation verified="yes">
 <filename>test.jpg</filename>
 <path>C:\Users\yasin\Desktop\test</path>
 <source>
 <database>Unknown</database>
 </source>
 <size>
 <width>400</width>
 <height>300</height>
 <depth>3</depth>
 </size>
 <segmented>0</segmented>
 <object>
 <name>People</name>
 <pose>Unspecified</pose>
 <truncated>0</truncated>
 <difficult>0</difficult>
 <bndbox>
  <xmin>80</xmin>
  <ymin>69</ymin>
  <xmax>144</xmax>
  <ymax>89</ymax>
 </bndbox>
 </object>
 <object>
 <name>CAT</name>
 <pose>Unspecified</pose>
 <truncated>0</truncated>
 <difficult>0</difficult>
 <bndbox>
  <xmin>40</xmin>
  <ymin>69</ymin>
  <xmax>143</xmax>
  <ymax>16</ymax>
 </bndbox>
 </object>
 <object>
 <name>dog</name>
 <pose>Unspecified</pose>
 <truncated>0</truncated>
 <difficult>0</difficult>
 <bndbox>
  <xmin>96</xmin>
  <ymin>82</ymin>
  <xmax>176</xmax>
  <ymax>87</ymax>
 </bndbox>
 </object> 
</annotation>

Python实现方案

具体实现

假设我们仅希望保留图像中的“people”和“cat”类别，并删除其他所有类别，可参考以下Python代码实现。


from xml.etree.ElementTree import ElementTree
from os import walk, path

def read_xml(in_path):
  tree = ElementTree()
  tree.parse(in_path)
  return tree

def write_xml(tree, out_path):
  tree.write(out_path, encoding="utf-8", xml_declaration=True)

def find_nodes(tree, path):
  return tree.findall(path)

def del_node_by_target_classes(nodelist, target_classes_lower, tree_root):
  for parent_node in nodelist:
    children = parent_node.getchildren()
    if (parent_node.tag == "object" and children[0].text.lower() not in target_classes_lower):
      tree_root.remove(parent_node)
    elif (parent_node.tag == "object" and children[0].text.lower() in target_classes_lower):
      children[0].text = children[0].text.lower()

def get_fileNames(rootdir):
  data_path = []
  prefixs = []
  for root, dirs, files in walk(rootdir, topdown=True):
    for name in files:
      pre, ending = path.splitext(name)
      if ending != ".xml":
        continue
      else:
        data_path.append(path.join(root, name))
        prefixs.append(pre)

  return data_path, prefixs

if __name__ == "__main__":
  # get all the xml paths, prefixes if not used here
  paths_xml, prefixs = get_fileNames("/home/yasin/old_labels/")

  target_classes = ["PEOPLE", "CAT"] # target flags you want to keep

  target_classes_lower = []
  for i in range(len(target_classes)):
    target_classes_lower.append(target_classes[i].lower()) # make sure your target is lowe-case

  # print(target_classes_lower)
  for i in range(len(paths_xml)):
    # rename and save the corresponding xml
    tree = read_xml(paths_xml[i])
    
    # get tree node
    tree_root = tree.getroot()

    # get parent nodes
    del_parent_nodes = find_nodes(tree, "./")
    
    # get target classes and delete
    target_del_node = del_node_by_target_classes(del_parent_nodes, target_classes_lower, tree_root)
    
    # save output xml, 000001.xml
    write_xml(tree, "/home/yasin/new_labels/{}.xml".format("%06d" % i))

执行上述代码后，原始XML文件将转换为新的XML格式。结果显示，除“people”和“cat”外的类别（如“dog”）已被移除，且保留类别的标签名称已统一转换为小写。


<?xml version='1.0' encoding='utf-8'?>
<annotation verified="yes">
 <filename>test.jpg</filename>
 <path>C:\Users\yasin\Desktop\test</path>
 <source>
 <database>Unknown</database>
 </source>
 <size>
 <width>400</width>
 <height>300</height>
 <depth>3</depth>
 </size>
 <segmented>0</segmented>
 <object>
 <name>people</name>
 <pose>Unspecified</pose>
 <truncated>0</truncated>
 <difficult>0</difficult>
 <bndbox>
  <xmin>80</xmin>
  <ymin>69</ymin>
  <xmax>144</xmax>
  <ymax>89</ymax>
 </bndbox>
 </object>
 <object>
 <name>cat</name>
 <pose>Unspecified</pose>
 <truncated>0</truncated>
 <difficult>0</difficult>
 <bndbox>
  <xmin>40</xmin>
  <ymin>69</ymin>
  <xmax>143</xmax>
  <ymax>16</ymax>
 </bndbox>
 </object>
</annotation>

以上即为本文的全部内容，希望对读者在数据预处理方面的学习有所帮助，也欢迎大家继续支持本站。