【目标检测】数据集准备全流程记录

土豆火腿包

已于 2023-02-03 16:12:04 修改

阅读量466

点赞数

文章标签：目标检测 python 深度学习

于 2023-02-03 16:07:30 首次发布

本文链接：https://blog.csdn.net/qq_33528632/article/details/128868464

版权

三种经典的数据集格式介绍

VOC格式

此处只以展示数据集中会用到的文件夹，数据集的格式如下：

VOC2007/
- Annotations/
  - 0000001.xml
  - 0000002.xml
  - …
- ImageSets/
  - Main
    - train.txt
    - test.txt
    - val.txt
- JPEGImages/
  - 0000001.jpg
  - 0000002.jpg
  - …

VOC格式的标签是xml格式，其中包含图片的诸多属性，示例如下：

<annotation>
	<folder>images</folder>
	<filename>000001.jpg</filename>
	<path>D:\360安全浏览器下载\Official-SSDD-OPEN\BBox_RBox_PSeg_SSDD\data_for_label\images\000001.jpg</path>
	<source>
		<database>Unknown</database>
	</source>
	<size>
		<width>416</width>
		<height>323</height>
		<depth>1</depth>
	</size>
	<segmented>0</segmented>
	<object>
		<name>ship</name>
		<pose>Unspecified</pose>
		<truncated>0</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>209</xmin>
			<ymin>44</ymin>
			<xmax>281</xmax>
			<ymax>148</ymax>
		</bndbox>
	</object>
</annotation>

其中<bndbox>表示框的gt值，是目标框的左上角和右下角的坐标。

COCO格式

coco/
- annotations/
  - instances_train2017.json
  - instances_val2017.json
- images/
  - train2017
    - 000001.jpg
    - …
  - val2017
    - 000002.jpg
    - …

数据集的标签均存放在json文件中，示例如下：

"annotations": [
{"area": 537.209371539205, 
 "iscrowd": 0,
 "image_id": 2, 
 "bbox": [211.51439299123905, 152.56570713391739, 49.56195244055067, 15.14392991239049], 
 "category_id": 0, 
 "id": 1, 
 "ignore": 0}]

其中"bbox"就是框的gt值，数值依次表示框的左上角坐标和宽高。

YOLO格式

dataset_name/
- images/
  - train
    - 000001.jpg
    - …
  - val
    - 000002.jpg
    - …
- labels/
  - train
    - 000001.txt
    - …
  - val
    - 000002.txt
    - …
- train.txt
- val.txt
- test.txt

YOLO数据集标签为txt格式，里面存放5个值：

0 0.469062 0.439437 0.115768 0.090141

分别表示：图像中框的类别，归一化后中心点的坐标，归一化后的宽高。
train.txt、val.txt以及test.txt 存放训练、验证和测试集图片的绝对路径，train.txt的示例如下：

D:/360安全浏览器下载/HRSID_JPG/images/train/P0001_0_800_7200_8000.jpg
D:/360安全浏览器下载/HRSID_JPG/images/train/P0001_0_800_8400_9200.jpg
D:/360安全浏览器下载/HRSID_JPG/images/train/P0001_0_800_9000_9800.jpg
D:/360安全浏览器下载/HRSID_JPG/images/train/P0001_0_800_9600_10400.jpg
D:/360安全浏览器下载/HRSID_JPG/images/train/P0001_1200_2000_0_800.jpg
D:/360安全浏览器下载/HRSID_JPG/images/train/P0001_1200_2000_4200_5000.jpg
D:/360安全浏览器下载/HRSID_JPG/images/train/P0001_1200_2000_7800_8600.jpg

数据集标注软件LabelImg安装

conda create -n labelimg python=3.6
conda activate labelimg
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple pyqt5
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple labelImg

最后在名为labelimg的虚拟环境中输入labelimg即可启动该标注软件。

数据集标注过程

Labelimg软件界面
可以直接选择左侧工具栏的Open Dir打开需要标注的图像文件夹路径，选择Change Save Dir确定好保存的标签的路径。VOC格式可创建文件夹名为Annotations;YOLO格式可创建文件夹名为labels。底下可以切换数据集的标注格式，有VOC、YOLO和ML格式。然后就可以选择Create Rectbox进行标注，标注完会弹出框让输入类别，最后Save即可！

VOC格式数据集划分

基于VOC格式划分，会在Main文件夹下生成train.txt、val.txt、test.txt三个txt文件，代码如下：

import os
import random
import argparse

parser = argparse.ArgumentParser()
# xml文件的地址，根据自己的数据进行修改 xml一般存放在Annotations下
parser.add_argument('--xml_path', default='Annotations/', type=str, help='input xml label path')
# 数据集的划分，地址选择自己数据下的ImageSets/Main
parser.add_argument('--txt_path', default='Main/', type=str, help='output txt label path')
opt = parser.parse_args()

train_percent = 0.8  # 训练集所占比例
val_percent = 0  # 验证集所占比例
test_persent = 0.2  # 测试集所占比例

xmlfilepath = opt.xml_path
txtsavepath = opt.txt_path
total_xml = os.listdir(xmlfilepath)

if not os.path.exists(txtsavepath):
    os.makedirs(txtsavepath)

num = len(total_xml)
list = list(range(num)) #list中存放除了train和val最后剩余的

t_train = int(num * train_percent)
t_val = int(num * val_percent)

train = random.sample(list, t_train) #随机选号
num1 = len(train)
for i in range(num1):
    list.remove(train[i])

val_test = [i for i in list if not i in train]
val = random.sample(val_test, t_val)
num2 = len(val)
for i in range(num2):
    list.remove(val[i])

file_train = open(txtsavepath + '/train.txt', 'w')
file_val = open(txtsavepath + '/val.txt', 'w')
file_test = open(txtsavepath + '/test.txt', 'w')

for i in train:
    name = total_xml[i][:-4] + '\n'
    file_train.write(name)

for i in val:
    name = total_xml[i][:-4] + '\n'
    file_val.write(name)

for i in list:
    name = total_xml[i][:-4] + '\n'
    file_test.write(name)

file_train.close()
file_val.close()
file_test.close()

运行示例：

python split_train_val.py --xml_path D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_voc_format/Annotations --txt_path D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_voc_format/Main

YOLO格式数据集划分

import os
import shutil
import random

random.seed(0)
def split_data(file_path, new_file_path, train_rate, val_rate, test_rate):
    eachclass_image = []
    for image in os.listdir(file_path):
        eachclass_image.append(image)
    total = len(eachclass_image)
    random.shuffle(eachclass_image)
    train_images = eachclass_image[0:int(train_rate * total)]
    val_images = eachclass_image[int(train_rate * total):int((train_rate + val_rate) * total)]
    test_images = eachclass_image[int((train_rate + val_rate) * total):]

    #创建三个txt 存放图像的绝对路径
    train_txt= open(new_file_path+'/train.txt', 'w', encoding='utf-8')
    if val_images:
        val_txt= open(new_file_path+'/val.txt', 'w', encoding='utf-8')
    if test_images:
        test_txt= open(new_file_path+'/test.txt', 'w', encoding='utf-8')

    #将训练集图片从原来的位置移动到新位置中的train文件夹
    for image in train_images:
        print(image)
        old_path = file_path + '/' + image
        new_path1 = new_file_path + '/' + 'images' + '/' + 'train'  #yolo/images/train
        if not os.path.exists(new_path1):
            os.makedirs(new_path1)
        new_path = new_path1 + '/' + image  #新图片存放位置
        shutil.copy(old_path, new_path)
        train_txt.write(new_path + '\n')

    new_name = os.listdir(new_file_path + '/' + 'images' + '/' + 'train')
    #移动label到新位置
    for im in new_name:
        old_xmlpath = xmlpath + '/' + im[:-3] + 'txt'
        new_xmlpath1 = new_file_path + '/' + 'labels' + '/' + 'train' #yolo/labels/train
        if not os.path.exists(new_xmlpath1):
            os.makedirs(new_xmlpath1)
        new_xmlpath = new_xmlpath1 + '/' + im[:-3] + 'txt'
        shutil.copy(old_xmlpath, new_xmlpath)

    for image in val_images:
        old_path = file_path + '/' + image
        new_path1 = new_file_path + '/' + 'images' + '/' + 'val'
        if not os.path.exists(new_path1):
            os.makedirs(new_path1)
        new_path = new_path1 + '/' + image
        shutil.copy(old_path, new_path)
        val_txt.write(new_path + '\n')

    if val_images:
        new_name = os.listdir(new_file_path + '/' + 'images' + '/' + 'val')

        for im in new_name:
            old_xmlpath = xmlpath + '/' + im[:-3] + 'txt'
            new_xmlpath1 = new_file_path + '/' + 'labels' + '/' + 'val'
            if not os.path.exists(new_xmlpath1):
                os.makedirs(new_xmlpath1)
            new_xmlpath = new_xmlpath1 + '/' + im[:-3] + 'txt'
            shutil.copy(old_xmlpath, new_xmlpath)

    for image in test_images:
        old_path = file_path + '/' + image
        new_path1 = new_file_path + '/' + 'images' + '/' + 'test'
        if not os.path.exists(new_path1):
            os.makedirs(new_path1)
        new_path = new_path1 + '/' + image
        shutil.copy(old_path, new_path)
        test_txt.write(new_path + '\n')
    if test_images:
        new_name = os.listdir(new_file_path + '/' + 'images' + '/' + 'test')

        for im in new_name:
            old_xmlpath = xmlpath + '/' + im[:-3] + 'txt'
            new_xmlpath1 = new_file_path + '/' + 'labels' + '/' + 'test'
            if not os.path.exists(new_xmlpath1):
                os.makedirs(new_xmlpath1)
            new_xmlpath = new_xmlpath1 + '/' + im[:-3] + 'txt'
            shutil.copy(old_xmlpath, new_xmlpath)

    train_txt.close()
    if val_images:
        val_txt.close()
    if test_images:
        test_txt.close()


if __name__ == '__main__':
    base_path= "D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/"
    file_path = base_path+'images'
    xmlpath = base_path+'labels'
    new_file_path = "D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/yolo"
    split_data(file_path, new_file_path, train_rate=0.8, val_rate=0.1, test_rate=0.1)

该代码会将图片和标签复制到新目录下，并创建train，test文件夹进行存放。

VOC转YOLO格式

import os
from os import getcwd
import xml.etree.ElementTree as ET

sets = ['train', 'test']
#这里使用要改成自己的类别
classes = ['ship']

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
    x = round(x, 6)
    w = round(w, 6)
    y = round(y, 6)
    h = round(h, 6)
    return x, y, w, h


# 后面只用修改各个文件夹的位置
def convert_annotation(image_id,image_set):
    # try:
    in_file = open('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_voc_format/Annotations/%s.xml' % (image_id), encoding='utf-8')
    out_file = open('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/labels/%s/%s.txt' % (image_set,image_id), 'w', encoding='utf-8')
    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
        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')

# except Exception as e:
# print(e, image_id)

wd = getcwd()
#创建yolo格式所需的labels文件夹，接着创建labels下的train和test文件夹
if not os.path.exists('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/labels/'):
    os.makedirs('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/labels/')

if not os.path.exists('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/labels/train/'):
    os.makedirs('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/labels/train/')
if not os.path.exists('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/labels/test/'):
    os.makedirs('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/labels/test/')

for image_set in sets:
    image_ids = open('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_voc_format/Main/%s.txt' %
                     (image_set)).read().strip().split()  #打开train.txt
    list_file = open('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/%s.txt' % (image_set), 'w')#最终要写入的txt文件
    for image_id in image_ids:
        print(image_id)
        list_file.write('D:/360安全浏览器下载/Official-SSDD-OPEN/BBox_RBox_PSeg_SSDD/data_for_label/data_yolo_format/images/%s/%s.jpg\n' % (image_set,image_id))
        convert_annotation(image_id,image_set)
    list_file.close()