1、前言：

本文演示如何划分数据集，以及将VOC标注的xml数据转为YOLO标注的txt格式，且生成classes的txt文件。

# 本文演示的项目目录
E:
└──dataset├── images_package  # 存放图片文件夹│    ├── 000002.jpg│    ├── 000003.jpg│    ├── 000004.jpg│    ├── 000005.jpg│    ├── 000006.jpg│    ├── zebra_crossing_20180129-000545_362.jpg│    ├── zebra_crossing_20180129-000545_364.jpg│    ├── zebra_crossing_20180129-000545_366.jpg│    └── zebra_crossing_20180129-000645_368.jpg└── xml_outputs  # 存放图片对应的XML├── 000002.xml├── 000003.xml├── 000004.xml├── 000005.xml├── 000006.xml├── zebra_crossing_20180129-000545_362.xml├── zebra_crossing_20180129-000545_364.xml├── zebra_crossing_20180129-000545_366.xml└── zebra_crossing_20180129-000645_368.xml

2、生成对应的类名

创建create_classes_json.py自动生成对应的类名json文件，以及在控制台输出对应的类名集。

from doctest import REPORTING_FLAGS
from lib2to3.pgen2.token import RPAR
import os
from tqdm import tqdm
from lxml import etree
import json# 读取 xml 文件信息，并返回字典形式
def parse_xml_to_dict(xml):if len(xml) == 0:  # 遍历到底层，直接返回 tag对应的信息return {xml.tag: xml.text}result = {}for child in xml:child_result = parse_xml_to_dict(child)  # 递归遍历标签信息if child.tag != 'object':result[child.tag] = child_result[child.tag]else:if child.tag not in result:  # 因为object可能有多个，所以需要放入列表里result[child.tag] = []result[child.tag].append(child_result[child.tag])return {xml.tag: result}# 提取xml中name保留为json文件
def xml2json(data,json_path):xml_path = [os.path.join(data, i) for i in os.listdir(data)]classes = []      # 目标类别num_object = 0for xml_file in tqdm(xml_path, desc="loading..."):with open(xml_file,encoding='gb18030',errors='ignore') as fid:      # 防止出现非法字符报错xml_str = fid.read()xml = etree.fromstring(xml_str)data = parse_xml_to_dict(xml)["annotation"]  # 读取xml文件信息for j in data['object']:        # 获取单个xml文件的目标信息ob = j['name']num_object +=1if ob not in classes:classes.append(ob)print(num_object)# 生成json文件labels = {}for index,object in enumerate(classes):labels[index] = object# 打印类名classes_name=[labels[key] for key in labels]print(f'类名：{classes_name}')# 打印类型字典print(f'字典形式：{labels}')# json.dumps将python对象转为json对象（将dict转化成str）。 json.loads将json字符串解码成python对象（将str转化成dict）labels = json.dumps(labels,indent=4)json_path=os.path.join(json_path,'classes_indices.json')with open(json_path,'w') as f:f.write(labels)if __name__ == "__main__":# 数据集的 xml 目录xml_path = 'E:\\dataset\\xml_outputs' # 存放类名json路径json_path='E:\\dataset'        xml2json(xml_path,json_path)pass'''
输出效果如下：
loading...: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [00:00<00:00, 143.24it/s]        
9
类名：['red', 'crosswalk']
字典形式：{0: 'red', 1: 'crosswalk'}
'''

classes_indices.json文件如下

3、xml转为yolo的label形式

创建xml_to_yolo_label_txt.py,在转换之前，我们需要获取数据集目标类别的列表，即运行上方create_classes_json.py，得到类名集，再替换CLASSES值。程序运行完后自动生成Annotations文件夹，里面就是yolo的label形式txt文件了。

import xml.etree.ElementTree as ET
import osdef convert(size,box):# 将bbox的左上角点，右下角点坐标的格式，转换为bbox中心点+bbox的W,H的格式，并进行归一化dw=1./size[0]dh=1./size[1]x=(box[0]+box[1])/2.0y=(box[2]+box[3])/2.0w=box[1]-box[0]h=box[3]-box[2]x=x*dww=w*dwy=y*dhh=h*dhreturn (x,y,w,h)def convert_annotation(xml_path,labels_path,CLASSES,image_id):# 把图像image_id的xml文件转换为目标检测的label文件(txt)# 其中包含物体的类别cls,bbox的中心点坐标,以及bbox的W,H# 并将四个物理量归一化in_file=open(xml_path+image_id)stats = os.stat(in_file.name)if stats.st_size!=0:image_id=image_id.split(".")[0]out_file=open(labels_path+"%s.txt"%(image_id),"w")tree=ET.parse(in_file)root=tree.getroot()size=root.find("size")w=int(size.find("width").text)h=int(size.find("height").text)for obj in root.iter("object"):#  difficult 代表是否难以识别，0表示易识别，1表示难识别。difficult=obj.find("difficult").textobj_cls=obj.find("name").textif obj_cls not in CLASSES:continuecls_id=CLASSES.index(obj_cls)xmlbox=obj.find("bndbox")points=(float(xmlbox.find("xmin").text),float(xmlbox.find("xmax").text),float(xmlbox.find("ymin").text),float(xmlbox.find("ymax").text))bb=convert((w,h),points)out_file.write(str(cls_id)+" "+" ".join([str(a) for a in bb])+"\n")def make_label_txt(xml_path,labels_path,CLASSES):# labels文件夹下创建image_id.txt# 对应每个image_id.xml提取出的bbox信息image_ids=os.listdir(xml_path)for file in image_ids:convert_annotation(xml_path,labels_path,CLASSES,file)if __name__=="__main__":# 类别，运行create_classes_json.py，得到类名集CLASSES=['red', 'crosswalk']# 数据整体文件路径common_path='E:\\dataset'# xml文件路径xml_path=os.path.join(common_path,'xml_outputs\\')# Annotations路径，即存放xml转为全部lables的路径labels_path=os.path.join(common_path,'Annotations\\')if not os.path.exists(labels_path):os.mkdir(labels_path)# 开始提取和转换make_label_txt(xml_path,labels_path,CLASSES)

4、优化代码

有人问，上方两个代码，需要运行两次，能否直接一次性运行就完呢？

答案：是可以的，上方代码关键就在获取类名集，于是将两者合并，创建final_xml2yolo.py代码

from doctest import REPORTING_FLAGS
from lib2to3.pgen2.token import RPAR
import os
from tqdm import tqdm
from lxml import etree
import json
import xml.etree.ElementTree as ET
import os# 读取 xml 文件信息，并返回字典形式
def parse_xml_to_dict(xml):if len(xml) == 0:  # 遍历到底层，直接返回 tag对应的信息return {xml.tag: xml.text}result = {}for child in xml:child_result = parse_xml_to_dict(child)  # 递归遍历标签信息if child.tag != 'object':result[child.tag] = child_result[child.tag]else:if child.tag not in result:  # 因为object可能有多个，所以需要放入列表里result[child.tag] = []result[child.tag].append(child_result[child.tag])return {xml.tag: result}# 提取xml中name保留为json文件
def xml2json(data,json_path):xml_path = [os.path.join(data, i) for i in os.listdir(data)]classes = []      # 目标类别num_object = 0for xml_file in tqdm(xml_path, desc="loading..."):with open(xml_file,encoding='gb18030',errors='ignore') as fid:      # 防止出现非法字符报错xml_str = fid.read()xml = etree.fromstring(xml_str)data = parse_xml_to_dict(xml)["annotation"]  # 读取xml文件信息for j in data['object']:        # 获取单个xml文件的目标信息ob = j['name']num_object +=1if ob not in classes:classes.append(ob)print(num_object)# 生成json文件labels = {}for index,object in enumerate(classes):labels[index] = object# 打印类名classes_name=[labels[key] for key in labels]print(f'类名：{classes_name}')# 打印类型字典print(f'字典形式：{labels}')# json.dumps将python对象转为json对象（将dict转化成str）。 json.loads将json字符串解码成python对象（将str转化成dict）labels = json.dumps(labels,indent=4)json_path=os.path.join(json_path,'classes_indices.json')with open(json_path,'w') as f:f.write(labels)# 返回类名 return classes_namedef convert(size,box):# 将bbox的左上角点，右下角点坐标的格式，转换为bbox中心点+bbox的W,H的格式，并进行归一化dw=1./size[0]dh=1./size[1]x=(box[0]+box[1])/2.0y=(box[2]+box[3])/2.0w=box[1]-box[0]h=box[3]-box[2]x=x*dww=w*dwy=y*dhh=h*dhreturn (x,y,w,h)def convert_annotation(xml_path,labels_path,CLASSES,image_id):# 把图像image_id的xml文件转换为目标检测的label文件(txt)# 其中包含物体的类别cls,bbox的中心点坐标,以及bbox的W,H# 并将四个物理量归一化in_file=open(xml_path+image_id)stats = os.stat(in_file.name)if stats.st_size!=0:image_id=image_id.split(".")[0]out_file=open(labels_path+"%s.txt"%(image_id),"w")tree=ET.parse(in_file)root=tree.getroot()size=root.find("size")w=int(size.find("width").text)h=int(size.find("height").text)for obj in root.iter("object"):#  difficult 代表是否难以识别，0表示易识别，1表示难识别。difficult=obj.find("difficult").textobj_cls=obj.find("name").textif obj_cls not in CLASSES:continuecls_id=CLASSES.index(obj_cls)xmlbox=obj.find("bndbox")points=(float(xmlbox.find("xmin").text),float(xmlbox.find("xmax").text),float(xmlbox.find("ymin").text),float(xmlbox.find("ymax").text))bb=convert((w,h),points)out_file.write(str(cls_id)+" "+" ".join([str(a) for a in bb])+"\n")def make_label_txt(xml_path,labels_path,CLASSES):# labels文件夹下创建image_id.txt# 对应每个image_id.xml提取出的bbox信息image_ids=os.listdir(xml_path)for file in image_ids:convert_annotation(xml_path,labels_path,CLASSES,file)if __name__ == "__main__":# 数据整体文件路径，同时也是存放类名json路径common_path=json_path='E:\\dataset'# xml文件路径xml_path=os.path.join(common_path,'xml_outputs\\')# 获取类别     CLASSES=xml2json(xml_path,json_path)# Annotations路径，即存放xml转为全部lables的路径labels_path=os.path.join(common_path,'Annotations\\')if not os.path.exists(labels_path):os.mkdir(labels_path)# 开始提取和转换make_label_txt(xml_path,labels_path,CLASSES)pass

5、划分数据集

创建split_train_val.py，根据具体情况分别修改cur_path,image_original_path,label_original_path 值，程序运行完后，分别生成images，labes和data_txt文件夹。

# 将图片和标注数据按比例切分为 训练集和测试集
import shutil
import random
import os# 原始路径
# 数据整体文件路径(注意这里需要修改)
cur_path='E:\\dataset'
# 图像文件夹路径,注意一定要有\\，(注意这里需要修改)
image_original_path = os.path.join(cur_path,"images_package\\")# 标注结果的路径即labels路径,该路径下不要有classes.txt,注意一定要有\\，(注意这里需要修改)
label_original_path = os.path.join(cur_path,"Annotations\\")		# cur_path = os.getcwd()# 训练集路径
train_image_path = os.path.join(cur_path, "images/train/")
train_label_path = os.path.join(cur_path, "labels/train/")# 验证集路径
val_image_path = os.path.join(cur_path, "images/val/")
val_label_path = os.path.join(cur_path, "labels/val/")# 测试集路径
test_image_path = os.path.join(cur_path, "images/test/")
test_label_path = os.path.join(cur_path, "labels/test/")# 训练集目录
data_txt_path=os.path.join(cur_path,'data_txt')
if not os.path.exists(data_txt_path):os.mkdir(data_txt_path)
list_train = os.path.join(data_txt_path, "train.txt")
list_val = os.path.join(data_txt_path, "val.txt")
list_test = os.path.join(data_txt_path, "test.txt")# 划分数据集比例
train_percent = 0.8
val_percent = 0.1
test_percent = 0.1def del_file(path):for i in os.listdir(path):file_data = path + "\\" + ios.remove(file_data)def mkdir():if not os.path.exists(train_image_path):os.makedirs(train_image_path)else:del_file(train_image_path)if not os.path.exists(train_label_path):os.makedirs(train_label_path)else:del_file(train_label_path)if not os.path.exists(val_image_path):os.makedirs(val_image_path)else:del_file(val_image_path)if not os.path.exists(val_label_path):os.makedirs(val_label_path)else:del_file(val_label_path)if not os.path.exists(test_image_path):os.makedirs(test_image_path)else:del_file(test_image_path)if not os.path.exists(test_label_path):os.makedirs(test_label_path)else:del_file(test_label_path)def clearfile():if os.path.exists(list_train):os.remove(list_train)if os.path.exists(list_val):os.remove(list_val)if os.path.exists(list_test):os.remove(list_test)def main():mkdir()clearfile()file_train = open(list_train, 'w')file_val = open(list_val, 'w')file_test = open(list_test, 'w')total_txt = os.listdir(label_original_path)num_txt = len(total_txt)list_all_txt = range(num_txt)num_train = int(num_txt * train_percent)num_val = int(num_txt * val_percent)num_test = num_txt - num_train - num_valtrain = random.sample(list_all_txt, num_train)# train从list_all_txt取出num_train个元素# 所以list_all_txt列表只剩下了这些元素val_test = [i for i in list_all_txt if not i in train]# 再从val_test取出num_val个元素，val_test剩下的元素就是testval = random.sample(val_test, num_val)print("训练集数目：{}, 验证集数目：{}, 测试集数目：{}".format(len(train), len(val), len(val_test) - len(val)))for i in list_all_txt:name = total_txt[i][:-4]srcImage = image_original_path + name + '.jpg'srcLabel = label_original_path + name + ".txt"if i in train:dst_train_Image = train_image_path + name + '.jpg'dst_train_Label = train_label_path + name + '.txt'shutil.copyfile(srcImage, dst_train_Image)shutil.copyfile(srcLabel, dst_train_Label)file_train.write(dst_train_Image + '\n')elif i in val:dst_val_Image = val_image_path + name + '.jpg'dst_val_Label = val_label_path + name + '.txt'shutil.copyfile(srcImage, dst_val_Image)shutil.copyfile(srcLabel, dst_val_Label)file_val.write(dst_val_Image + '\n')else:dst_test_Image = test_image_path + name + '.jpg'dst_test_Label = test_label_path + name + '.txt'shutil.copyfile(srcImage, dst_test_Image)shutil.copyfile(srcLabel, dst_test_Label)file_test.write(dst_test_Image + '\n')file_train.close()file_val.close()file_test.close()if __name__ == "__main__":main()

本文最终形成的目录形式

    E:└──dataset├── Annotations│    ├── 000002.txt│    ├── 000003.txt│    ├── 000004.txt│    ├── 000005.txt│    ├── 000006.txt│    ├── zebra_crossing_20180129-000545_362.txt│    ├── zebra_crossing_20180129-000545_364.txt│    ├── zebra_crossing_20180129-000545_366.txt│    └── zebra_crossing_20180129-000645_368.txt├── classes_indices.json├── data_txt│    ├── test.txt│    ├── train.txt│    └── val.txt├── images│    ├── test│    │    ├── 000005.jpg│    │    └── zebra_crossing_20180129-000545_362.jpg│    ├── train│    │    ├── 000002.jpg│    │    ├── 000003.jpg│    │    ├── 000004.jpg│    │    ├── 000006.jpg│    │    ├── zebra_crossing_20180129-000545_364.jpg│    │    ├── zebra_crossing_20180129-000545_366.jpg│    │    └── zebra_crossing_20180129-000645_368.jpg│    └── val├── images_package│    ├── 000002.jpg│    ├── 000003.jpg│    ├── 000004.jpg│    ├── 000005.jpg│    ├── 000006.jpg│    ├── zebra_crossing_20180129-000545_362.jpg│    ├── zebra_crossing_20180129-000545_364.jpg│    ├── zebra_crossing_20180129-000545_366.jpg│    └── zebra_crossing_20180129-000645_368.jpg├── labels│    ├── test│    │    ├── 000005.txt│    │    └── zebra_crossing_20180129-000545_362.txt│    ├── train│    │    ├── 000002.txt│    │    ├── 000003.txt│    │    ├── 000004.txt│    │    ├── 000006.txt│    │    ├── zebra_crossing_20180129-000545_364.txt│    │    ├── zebra_crossing_20180129-000545_366.txt│    │    └── zebra_crossing_20180129-000645_368.txt│    └── val└── xml_outputs├── 000002.xml├── 000003.xml├── 000004.xml├── 000005.xml├── 000006.xml├── zebra_crossing_20180129-000545_362.xml├── zebra_crossing_20180129-000545_364.xml├── zebra_crossing_20180129-000545_366.xml└── zebra_crossing_20180129-000645_368.xml

6、画目录树

创建draw_tree.py，如本文输入

请输入文件夹路径（不含名称）： E

请输入文件夹名称：dataset

自动生成 tree.txt

import osdef get_num(path):dirlist = os.listdir(path)j=0for i in dirlist:j+=1return jdef print_tree(path,last):num=get_num(path)if num!=0:dirlist = os.listdir(path)j=0for i in dirlist:for k in last:if k=='0':print("  │",end=" ")else:print("   ", end=" ")j+=1if j<num:print("  ├── ", end="")print(i)dir=path+"\\"+iif os.path.isdir(dir):print_tree(dir,last+'0')else:print("  └── ", end="")print(i)dir = path + "\\" + iif os.path.isdir(dir):print_tree(dir,last+'1')def write_tree(path,last,f):num=get_num(path)if num!=0:dirlist = os.listdir(path)j=0for i in dirlist:for k in last:if k=='0':f.write("    │")else:f.write("    ")j+=1if j<num:f.write("    ├── ")f.write(i)f.write('\n')dir=path+"\\"+iif os.path.isdir(dir):write_tree(dir,last+'0',f)else:f.write("    └── ")f.write(i)f.write('\n')dir = path + "\\" + iif os.path.isdir(dir):write_tree(dir,last+'1',f)if __name__=='__main__':path = input("请输入文件夹路径（不含名称）：")root = input("请输入文件夹名称：")if len(path)==1:path+=':'#print("  └─root")#print_tree('D:\\root',"1")f = open("tree.txt", "w", encoding="utf-8")f.write("    └──"+root+"\n")write_tree(path+"\\"+root, "1",f)f.close()

7、目标检测系列文章

1. YOLOv5s网络模型讲解(一看就会)

2. 生活垃圾数据集（YOLO版）

3. YOLOv5如何训练自己的数据集

4. 双向控制舵机（树莓派版）

5. 树莓派部署YOLOv5目标检测（详细篇）

6. YOLO_Tracking 实践 （环境搭建 & 案例测试）