Yolov5训练自己的数据集(最详细教程)

一、环境配置部分

默认使用anaconda来管理python环境。

1.创建虚拟环境

conda create -n yolov5 python=3.8

2.根据自己安装的CUDA版本去pytorch官网下载torch等。

因为我的CUDA是11.1，默认的没有，点击下面的previous versions of pytorch看以前的版本。发现torch1.10.1可以

pip install torch==1.10.1+cu111 torchvision==0.11.2+cu111 torchaudio==0.10.1 -f https://download.pytorch.org/whl/cu111/torch_stable.html

3.下载yolov5代码

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

二、数据集制作部分

先用labelimage标注自己的数据，尽量是yolo格式的，也就是标注文件是txt形式，如果有voc格式也没关系，可以进行转换。

2.1voc格式

?第一步：把JPEGImages、Annotations、ImageSets(他们的说明见下面代码)和下面的代码放在一个目录下。运行下面的代码，就会在ImageSets/Main/下得到训练和验证的txt文件。

import os
import random

images_path = "JPEGImages/"       #里面放的图片
xmls_path = "Annotations/"        #里面放的xml格式标注文件
train_val_txt_path = "ImageSets/Main/"   #这个就是一个空的文件夹，运行这个代码后在Main文件夹下有两个训练的txt文件
val_percent = 0.1     #验证集的比例。

images_list = os.listdir(images_path)
random.shuffle(images_list)

train_images_count = int((1 - val_percent) * len(images_list))
val_images_count = int(val_percent * len(images_list))

train_txt = open(os.path.join(train_val_txt_path, "train.txt"), "w")
train_count = 0
for i in range(train_images_count):
    text = images_list[i].split(".png")[0] + "\n"
    train_txt.write(text)
    train_count += 1
    print("train_count : " + str(train_count))
train_txt.close()

val_txt = open(os.path.join(train_val_txt_path, "val.txt"), "w")
val_count = 0
for i in range(val_images_count):
    text = images_list[i + train_images_count].split(".png")[0] + "\n"
    val_txt.write(text)
    val_count += 1
    print("val_count : " + str(val_count))
val_txt.close()

第二步：标注文件voc格式转yolo格式

我们只需要在main函数里指定两个参数，一个是VOC标注文件地址，一个是yolo输出文件地址

import os
import xml.etree.ElementTree as ET
 
def convert_folder_to_yolov5(input_folder, output_folder):
    # Ensure output folder exists
    if not os.path.exists(output_folder):
        os.makedirs(output_folder)
 
    # Loop through each XML file in the input folder
    for xml_file_name in os.listdir(input_folder):
        if xml_file_name.endswith('.xml'):
            xml_file_path = os.path.join(input_folder, xml_file_name)
 
            # Generate corresponding output txt file path
            txt_file_name = os.path.splitext(xml_file_name)[0] + '.txt'
            txt_file_path = os.path.join(output_folder, txt_file_name)
 
            # Convert XML to Yolov5 format and save to txt file
            convert_to_yolov5(xml_file_path, txt_file_path)
 
def convert_to_yolov5(xml_file, output_file):
    tree = ET.parse(xml_file)
    root = tree.getroot()
 
    with open(output_file, 'w') as f:
        for obj in root.findall('object'):
            class_name = obj.find('name').text
            if class_name == 'cone':  # Assuming 'disease' is the class of interest
                xmin = int(obj.find('bndbox/xmin').text)
                ymin = int(obj.find('bndbox/ymin').text)
                xmax = int(obj.find('bndbox/xmax').text)
                ymax = int(obj.find('bndbox/ymax').text)
 
                width = xmax - xmin
                height = ymax - ymin
                x_center = (xmin + xmax) / 2.0
                y_center = (ymin + ymax) / 2.0
 
                # Normalize coordinates and dimensions
                x_center /= int(root.find('size/width').text)
                y_center /= int(root.find('size/height').text)
                width /= int(root.find('size/width').text)
                height /= int(root.find('size/height').text)
 
                line = f"{0} {x_center} {y_center} {width} {height}\n"
                f.write(line)
 
if __name__ == "__main__":
    input_folder_path = "/home/wangchen/YOLOX/cone/Annotations"  #voc格式标注文件
    output_folder_path = "/home/wangchen/YOLOX/cone/YOLOLabels"   #yolo格式保存地址
 
    convert_folder_to_yolov5(input_folder_path, output_folder_path)

第三步：根据第一步的生成的voc索引，来将yolo数据划分为train和val两部分。

下面这个代码运行完，会在output_dataset_path里面产生两个文件夹，一个train，一个val，每一个里面又都有一个images和labels。这个数据集目录结构是不对的，需要调整一下。改成下图结构

import os
import random
from shutil import copyfile


def split_dataset(image_folder, txt_folder, output_folder, split_index):
    # Ensure output folders exist
    for dataset in ['train', 'val']:
        if not os.path.exists(os.path.join(output_folder, dataset, 'images')):
            os.makedirs(os.path.join(output_folder, dataset, 'images'))
        if not os.path.exists(os.path.join(output_folder, dataset, 'txt')):
            os.makedirs(os.path.join(output_folder, dataset, 'txt'))
    train_index = os.path.join(split_index, 'train.txt')
    val_index = os.path.join(split_index, 'val.txt')
    with open(train_index, 'r') as file:
        train_images = [i.strip() for i in file.readlines()]
    with open(val_index, 'r') as file:
        val_images = [i.strip() for i in file.readlines()]

    # Copy images to respective folders
    for dataset, images_list in zip(['train', 'val'], [train_images, val_images]):
        for image_file in images_list:
            image_path = os.path.join(image_folder, image_file + '.png')
            copyfile(image_path, os.path.join(output_folder, dataset, 'images', image_file + '.png'))
            txt_file = image_file + '.txt'
            txt_path = os.path.join(txt_folder, txt_file)

            # Copy corresponding txt file if exists
            if os.path.exists(txt_path):
                copyfile(txt_path, os.path.join(output_folder, dataset, 'txt', txt_file))


if __name__ == "__main__":
    image_folder_path = "/home/wangchen/YOLOX/cone/JPEGImages"
    txt_folder_path = "/home/wangchen/YOLOX/cone/YOLOLabels"
    output_dataset_path = "/home/wangchen/YOLOX/yolo_data"
    split_index = "/home/wangchen/YOLOX/cone/ImageSets/Main"

    split_dataset(image_folder_path, txt_folder_path, output_dataset_path, split_index)

2.2YOLO格式

直接按照上面第三步目录结构划分就行。

三、yolov5配置文件修改

?修改data/VOC.yaml.

修改models/yolov5_s.yaml里面的类别个数。修改train.py里面的相关超参数即可。