yolov5训练自己数据集流程

本文将介绍yolov5模型训练的整个过程。

二：yolov5训练

1.官网下载源代码：https://github.com/ultralytics/yolov5

2.使用pycharm打开解压后的yolov5文件夹

3.打开主文件夹目录下的train.py文件，对参数进行修改

4.使用labelimg进行标注，建议标签名不要用中文。

标注完得到xml文件后，使用脚本进行转化文件格式为txt。

修改xml_path的文件路径（换成自己的），再运行以下代码 。

#!/usr/bin/env python3

# -*- coding: utf-8 -*-

import copy

from lxml.etree import Element, SubElement, tostring, ElementTree

import xml.etree.ElementTree as ET

import pickle

import os

from os import listdir, getcwd

from os.path import join

classes = ["0", "1", "2", "3"]  # 类别

CURRENT_DIR = os.path.dirname(os.path.abspath(__file__))

def convert(size, box):

    dw = 1. / size[0]

    dh = 1. / size[1]

    x = (box[0] + box[1]) / 2.0

    y = (box[2] + box[3]) / 2.0

    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('./label_xml\%s.xml' % (image_id), encoding='UTF-8')

    out_file = open('./label_txt\%s.txt' % (image_id), 'w')  # 生成txt格式文件

    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'):

        cls = obj.find('name').text

        # print(cls)

        if cls not in classes:

            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))

        bb = convert((w, h), b)

        out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')

xml_path = os.path.join(CURRENT_DIR, './label_xml/')

# xml list

img_xmls = os.listdir(xml_path)

for img_xml in img_xmls:

    label_name = img_xml.split('.')[0]

    print(label_name)

    convert_annotation(label_name)

5.创建文件夹mydata,将对应xml、img/bmp、txt文件分别存入对应文件夹中。

6.在mydata文件夹路径下创建split.py文件，使用以下代码对txt以及img/bmp文件进行训练集、验证集、测试集划分，修改 img_path、label_path为自己的文件路径。

'''

Descripttion: split_img.py

version: 1.0

Author: UniDome

Date: 2022-04-20 16:28:45

LastEditors: UniDome

LastEditTime: 2022-04-20 16:39:56

'''

import os, shutil, random

from tqdm import tqdm

def split_img(img_path, label_path, split_list):

    try :   # 创建数据集文件夹

        Data = '../datasets'

        os.mkdir(Data)

        train_img_dir = Data + '/images/train'

        val_img_dir = Data + '/images/val'

        test_img_dir = Data + '/images/test'

        train_label_dir = Data + '/labels/train'

        val_label_dir = Data + '/labels/val'

        test_label_dir = Data + '/labels/test'

        # 创建文件夹

        os.makedirs(train_img_dir)

        os.makedirs(train_label_dir)

        os.makedirs(val_img_dir)

        os.makedirs(val_label_dir)

        os.makedirs(test_img_dir)

        os.makedirs(test_label_dir)

    except:

        print('文件目录已存在')

       

    train, val, test = split_list

    all_img = os.listdir(img_path)

    all_img_path = [os.path.join(img_path, img) for img in all_img]

    # all_label = os.listdir(label_path)

    # all_label_path = [os.path.join(label_path, label) for label in all_label]

    train_img = random.sample(all_img_path, int(train * len(all_img_path)))

    train_img_copy = [os.path.join(train_img_dir, img.split('\\')[-1]) for img in train_img]

    train_label = [toLabelPath(img, label_path) for img in train_img]

    train_label_copy = [os.path.join(train_label_dir, label.split('\\')[-1]) for label in train_label]

    for i in tqdm(range(len(train_img)), desc='train ', ncols=80, unit='img'):

        _copy(train_img[i], train_img_dir)

        _copy(train_label[i], train_label_dir)

        all_img_path.remove(train_img[i])

    val_img = random.sample(all_img_path, int(val / (val + test) * len(all_img_path)))

    val_label = [toLabelPath(img, label_path) for img in val_img]

    for i in tqdm(range(len(val_img)), desc='val ', ncols=80, unit='img'):

        _copy(val_img[i], val_img_dir)

        _copy(val_label[i], val_label_dir)

        all_img_path.remove(val_img[i])

    test_img = all_img_path

    test_label = [toLabelPath(img, label_path) for img in test_img]

    for i in tqdm(range(len(test_img)), desc='test ', ncols=80, unit='img'):

        _copy(test_img[i], test_img_dir)

        _copy(test_label[i], test_label_dir)

def _copy(from_path, to_path):

    shutil.copy(from_path, to_path)

def toLabelPath(img_path, label_path):

    img = img_path.split('\\')[-1]

    label = img.split('.jpg')[0] + '.txt'

    return os.path.join(label_path, label)

def main():

    img_path = r'D:\malong\python\yolov5-7.0\mydata\images'

    label_path = r'D:\malong\python\yolov5-7.0\mydata\labels'

    split_list = [0.7, 0.2, 0.1]        # 数据集划分比例[train:val:test]

    split_img(img_path, label_path, split_list)

if __name__ == '__main__':

    main()

7.前文第三点所说的coco128.yaml文件，修改为自己需要的数据集格式。

8.设置好之后运行train.py文件，出现以下界面则训练成功，等待完成训练即可。