模型训练时的互转标注文件格式以及标注验证代码（标注文件格式转换和标注验证）

m0_56843848

已于 2023-10-19 04:29:25 修改

阅读量124

点赞数

分类专栏：便捷工具文章标签：笔记

于 2023-10-12 04:11:42 首次发布

本文链接：https://blog.csdn.net/m0_56843848/article/details/133782230

版权

便捷工具专栏收录该内容

15 篇文章 0 订阅

订阅专栏

2、YOLO格式转到XML格式：.txt→.xml

二、标注验证

1、XML格式文件的验证xml

2、YOLO格式文件的验证txt

一、标注文件格式转换

1、XML格式转到YOLO格式：.xml→.txt

我觉得第一个好

第1个xml_to_labels.py

#    xml_to_labels.py

from xml.etree import ElementTree as ET
import numpy as np
import os
import shutil
from pathlib import Path
from shutil import copyfile
from PIL import Image, ImageDraw
from xml.dom.minidom import parse
from tqdm import tqdm

"""
修改类别，labels文件路径，图片文件路径，annotation文件路径
"""


def cord_converter(size, box):
    """
    将标注的 xml 文件标注转换为 darknet 形的坐标
    :param size: 图片的尺寸： [w,h]
    :param box: anchor box 的坐标 [左上角x,左上角y,右下角x,右下角y,]
    :return: 转换后的 [x,y,w,h]
    """

    x1 = int(box[0])
    y1 = int(box[1])
    x2 = int(box[2])
    y2 = int(box[3])

    dw = np.float32(1. / int(size[0]))
    dh = np.float32(1. / int(size[1]))

    w = x2 - x1
    h = y2 - y1
    x = x1 + (w / 2)
    y = y1 + (h / 2)

    return [x, y, w, h]


class XMLProcess(object):

    def __init__(self, num_classes, classes_name, file_path, labels_path):
        self.xml_path = file_path
        self.num_classes = num_classes
        self.classes_name = classes_name
        self.data = {}
        self.labels_path = labels_path

    def process_xml(self):
        """
        处理图片的标注信息，解析图片的大小，类别。并存入.pkl文件
        :return: None
        """
        i = 1
        for filename in tqdm(os.listdir(self.xml_path)):

            # xml解析根路径
            et = ET.parse(self.xml_path + filename)
            root = et.getroot()

            # 获取图片基础属性 size ---> [width, height, depth]
            size = root.find('size')
            width = float(size.find("width").text)
            height = float(size.find("height").text)
            depth = float(size.find("depth").text)
            # print(width, height, depth)

            # 解析每张图片中多个物体的属性
            object_labels = []
            one_hots = []
            for object_tree in root.findall("object"):
                # 对每个目标进行解析
                [x, y, w, h] = [0, 0, 0, 0]
                for res in object_tree.iter("bndbox"):
                    # xmin, ymin, xmax, ymax---> 以标准化的结果去保存值
                    xmin = float(res.find("xmin").text) / width
                    ymin = float(res.find("ymin").text) / height
                    xmax = float(res.find("xmax").text) / width
                    ymax = float(res.find("ymax").text) / height
                    # print(xmin, ymin, xmax, ymax)
                    # 将[x1, y1, x2, y2] 转换成 [x, y, ,w, h]
                    w = float(xmax - xmin)
                    h = float(ymax - ymin)
                    x = float(xmin + w / 2)
                    y = float(ymin + h / 2)

                # 每个object都有一个名称，目标值保存为one-hot编码
                object_name = object_tree.find('name').text
                object_one_hot = self.one_hot(object_name)
                object_id = str(np.argmax(object_one_hot))  # 目标类别编号
                # print([object_id, x, y, w, h])
                object_label = [object_id, x, y, w, h]
                object_labels.append(object_label)
            # print(object_labels)

            # 创建Labels文件夹
            labels_path = self.labels_path
            if not os.path.exists(labels_path):
                os.makedirs(labels_path)

            # 将labels写入.txt文件
            # write_path = f"../datasets/part_yawn/Labels/{filename[:-4]}.txt"
            write_path = f"{labels_path}/{filename[:-4]}.txt"

            # write_path = f'../commodity_label/{filename[:-4]}.txt'
            f = open(write_path, 'w')
            for label in object_labels:
                for item in label:
                    f.write(str(item) + ' ')
                f.write("\n")
            if i == 4:
                break
        return None

    def one_hot(self, name):
        """
        one-hot编码
        :param name: 目标名称
        :return: one_hot_vector
        """
        one_hot_vector = [0] * self.num_classes
        classes_name = self.classes_name
        identify_num = 0
        for i, commodity_name in enumerate(classes_name):
            if name == classes_name[i]:
                one_hot_vector[i] = 1
                break
            else:
                identify_num += 1
                # print(identify_num)
                if identify_num >= self.num_classes:
                    print("Unknown Label: ", name)
        return one_hot_vector


if __name__ == '__main__':
    # a = XMLProcess(num_classes=4,
    #                classes_name=['fire', 'smoke'],
    #                file_path=r'D:\桌面文件\fix\anno\\',       # xml文件标注路径  old
    #                labels_path=r'D:\桌面文件\fix\labels')             # txt文件路径       new
    #
    a = XMLProcess(num_classes=4,
                   classes_name=['open_mouth', 'closed_mouth', 'open_eye', 'closed_eye'],               # 依次为0，1，2，3
                   file_path=r'F:\yolo\yolov8\ultralytics-main\datasets\fdd-dataset\Annotations\\',       # xml文件标注路径  old
                   labels_path=r'F:\yolo\yolov8\ultralytics-main\datasets\fdd-dataset\labels\\')             # txt文件路径       new
    a.process_xml()

第2个xml_to_labels-2.py

#    xml_to_labels-2.py

# -*- coding: utf-8 -*-
import xml.etree.ElementTree as ET
import os
 
sets = ['train', 'val', 'test']  # 如果你的Main文件夹没有test.txt，就删掉'test'
# classes = ["a", "b"]   # 改成自己的类别，VOC数据集有以下20类别
classes = ['person']  # class names
abs_path = os.getcwd()
 
 
def convert(size, box):
    dw = 1. / (size[0])
    dh = 1. / (size[1])
    x = (box[0] + box[1]) / 2.0 - 1
    y = (box[2] + box[3]) / 2.0 - 1
    w = box[1] - box[0]
    h = box[3] - box[2]
    x = x * dw
    w = w * dw
    y = y * dh
    h = h * dh
    return x, y, w, h
 
 
def convert_annotation(image_id):
    in_file = open(abs_path + '/INRIAPerson/VOCperson/Annotations/%s.xml' % (image_id))
    out_file = open(abs_path + '/INRIAPerson/VOCperson/label/%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 = obj.find('difficult').text
        # difficult = obj.find('Difficult').text
        cls = obj.find('name').text
        if cls not in classes or int(difficult) == 1:
            continue
        cls_id = classes.index(cls)
        xmlbox = obj.find('bndbox')
        b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text), float(xmlbox.find('ymin').text),
             float(xmlbox.find('ymax').text))
        b1, b2, b3, b4 = b
        # 标注越界修正
        if b2 > w:
            b2 = w
        if b4 > h:
            b4 = h
        b = (b1, b2, b3, b4)
        bb = convert((w, h), b)
        out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')
 
 
for image_set in sets:
    if not os.path.exists(abs_path + '/INRIAPerson/VOCperson/label/'):
        os.makedirs(abs_path + '/INRIAPerson/VOCperson/label/')
 
    image_ids = open(abs_path + '/INRIAPerson/VOCperson/ImageSets/Main/%s.txt' % (image_set)).read().strip().split()
    list_file = open(abs_path + '/INRIAPerson/VOCperson/VOC2007/%s.txt' % (image_set), 'w')#文件输出的路径
    for image_id in image_ids:
        list_file.write(abs_path + '/INRIAPerson/VOCperson/JPEGImages/%s.jpg\n' % (image_id))  # 要么自己补全路径，只写一半可能会报错
        convert_annotation(image_id)
    list_file.close()

第二个实现两个功能（1）XML格式标注文件转换到YOLO格式标注文件

（2）产生train.txt、val.txt以及test.txt文件的路径形式

下面代码只实现（2）的功能

数据集划分txt.py

#数据集划分txt.py
# -*- coding:utf-8 -*
import os
import random

val_percent = 0.1
test_percent = 0.1
train_percent = 0.8
imagesfile_path  = './images/'    # 图片名
total_images = os.listdir(imagesfile_path)
random.shuffle(total_images)
num = len(total_images)  # 统计所有的标注文件
train_num = int(num * train_percent)  # 设置训练和验证集的数目
val_num = int(num * val_percent)  # 设置训练集的数目
test_num = int(num * test_percent)
ftest = open('./test.txt', 'w')
ftrain = open('./train.txt', 'w')
fval = open('./val.txt', 'w')
for i in range(num):
    # name = './images/'+total_images[i] + '\n'
    name = './images/'+total_images[i] + '\n'

    if i < train_num:
        ftrain.write(name)
    elif train_num <= i < val_num + train_num:
        fval.write(name)
    else:
        ftest.write(name)
ftrain.close()
fval.close()
ftest.close()

数据集划分txt2.py

#数据集划分txt2.py
# -*- coding: utf-8 -*-
import xml.etree.ElementTree as ET
import os
 
sets = ['train', 'val','test']  # 如果你的Main文件夹没有test.txt，就删掉'test'
# classes = ["a", "b"]   # 改成自己的类别，VOC数据集有以下20类别
classes = ['person']  # class names
abs_path = os.getcwd()
 
for image_set in sets:
 
 
    image_ids = open(abs_path + '/KAIST/ImageSets/Main/%s.txt' % (image_set)).read().strip().split()
    list_file = open(abs_path + '/KAIST/%s.txt' % (image_set), 'w')#文件输出的路径
    for image_id in image_ids:
        list_file.write(abs_path + '/KAIST/Images/%s.jpg\n' % (image_id))  # 要么自己补全路径，只写一半可能会报错
 
    list_file.close()