标注数据集处理---Python制作VOC标注格式的xml标注文件
引:
近期做CV方面算法,分享几个简单的视频、图片处理脚本
脚本中均有print调试代码,,方便更改
Python制作VOC格式xml的函数方法:
import os
import xml.dom.minidom
def write_xml(folder: str, img_name: str, path: str, img_width: int, img_height: int, tag_num: int, tag_name: str, box_list:list):
'''
VOC标注xml文件生成函数
:param folder: 文件夹名
:param img_name:
:param path:
:param img_width:
:param img_height:
:param tag_num: 图片内的标注框数量
:param tag_name: 标注名称
:param box_list: 标注坐标,其数据格式为[[xmin1, ymin1, xmax1, ymax1],[xmin2, ymin2, xmax2, ymax2]....]
:return: a standard VOC format .xml file, named "img_name.xml"
'''
# 创建dom树对象
doc = xml.dom.minidom.Document()
# 创建root结点annotation,并用dom对象添加根结点
root_node = doc.createElement("annotation")
doc.appendChild(root_node)
# 创建结点并加入到根结点
folder_node = doc.createElement("folder")
folder_value = doc.createTextNode(folder)
folder_node.appendChild(folder_value)
root_node.appendChild(folder_node)
filename_node = doc.createElement("filename")
filename_value = doc.createTextNode(img_name)
filename_node.appendChild(filename_value)
root_node.appendChild(filename_node)
path_node = doc.createElement("path")
path_value = doc.createTextNode(path)
path_node.appendChild(path_value)
root_node.appendChild(path_node)
source_node = doc.createElement("source")
database_node = doc.createElement("database")
database_node.appendChild(doc.createTextNode("Unknown"))
source_node.appendChild(database_node)
root_node.appendChild(source_node)
size_node = doc.createElement("size")
for item, value in zip(["width", "height", "depth"], [img_width, img_height, 3]):
elem = doc.createElement(item)
elem.appendChild(doc.createTextNode(str(value)))
size_node.appendChild(elem)
root_node.appendChild(size_node)
seg_node = doc.createElement("segmented")
seg_node.appendChild(doc.createTextNode(str(0)))
root_node.appendChild(seg_node)
for i in range(tag_num):
obj_node = doc.createElement("object")
name_node = doc.createElement("name")
name_node.appendChild(doc.createTextNode(tag_name))
obj_node.appendChild(name_node)
pose_node = doc.createElement("pose")
pose_node.appendChild(doc.createTextNode("Unspecified"))
obj_node.appendChild(pose_node)
trun_node = doc.createElement("truncated")
trun_node.appendChild(doc.createTextNode(str(0)))
obj_node.appendChild(trun_node)
trun_node = doc.createElement("difficult")
trun_node.appendChild(doc.createTextNode(str(0)))
obj_node.appendChild(trun_node)
bndbox_node = doc.createElement("bndbox")
for item, value in zip(["xmin", "ymin", "xmax", "ymax"], box_list[i]):
elem = doc.createElement(item)
elem.appendChild(doc.createTextNode(str(value)))
bndbox_node.appendChild(elem)
obj_node.appendChild(bndbox_node)
root_node.appendChild(obj_node)
with open(img_name.split('.')[-2] + ".xml", "w", encoding="utf-8") as f:
# writexml()第一个参数是目标文件对象,第二个参数是根节点的缩进格式,第三个参数是其他子节点的缩进格式,
# 第四个参数制定了换行格式,第五个参数制定了xml内容的编码。
doc.writexml(f, indent='', addindent='\t', newl='\n', encoding="utf-8")
方法使用演示(以lp-annot.idl为例) :
if __name__ == '__main__':
f = open("./lp-annot.idl", "r")
lines = f.readlines() # 读取全部内容 ,并以列表方式返回
for line in lines:
try:
line_list = line.split(":-1")
# print(f'line_list: {line_list}\nlen(line_list):{len(line_list)}\n')
temp_line0_file_name = line_list[0].split(':')[0].split('"')[1].split('/')[-1]
temp_line0_tag = line_list[0].split(':')[1]
print(f'temp_line0_file_name: {temp_line0_file_name} temp_line0_tag: {temp_line0_tag}')
new_line_list = list()
new_line_list.append(temp_line0_tag)
for i in range(1, len(line_list) - 1):
new_line_list.append(line_list[i])
print(f'new_line_list: {new_line_list}')
box_list = []
for i in range(len(new_line_list)):
print(f'new_line_list[i]: {new_line_list[i]}')
box = new_line_list[i].split("(")[1].split(',')
box[3] = box[3].split(')')[0]
print(f"box: {box}")
x1, y1, x2, y2 = box[0], box[1], box[2], box[3]
# if (int(x2) - int(x1)) >= (50*(float(img_width)/640.0)) and (int(y2) - int(y1)) >= (50*(float(img_height)/640.0)):
if (int(x2) - int(x1)) >= 40 and (int(y2) - int(y1)) >= 70:
box_list.append(box[:])
print(f"box_list: {box_list}")
if len(box_list) == 0:
continue
write_xml(folder='VOC2014_instance/person', img_name='lp-annot_' + temp_line0_file_name,
path=temp_line0_file_name, img_width=640, img_height=480, tag_num=len(box_list),
tag_name='person', box_list=box_list)
except Exception as e:
print("ERROR, e----------------------------------------------\n:", e, "\n--------------------------------")

浙公网安备 33010602011771号