二万字史诗保姆级远程服务器YOLOv5制作训练自己数据集！

本文链接：https://blog.csdn.net/m0_65619221/article/details/131986265

1.远程服务器搭建环境，连接服务器

1.1下载yolov5代码：

GitHub - ultralytics/yolov5: YOLOv5 🚀 in PyTorch > ONNX > CoreML > TFLite

服务器租用、资料上传、环境搭建、VNC操作教程之类的网上都有很简单可以自查一下。

1.2我这里采用的矩池云，客服挺好的，而且里面有yolov5的环境可以基本直接使用。

1.3连接方式建议VNC直接远程操控，网盘都在/mnt文件夹下，比用pychrm操作简单方便。

下载地址：https://downloads.realvnc.com/download/file/viewer.files/VNC-Viewer-7.1.0-Windows.exe

2.相机拍摄制作目标多姿态数据集。

将图片顺序化重复名以便数据增强。（关注私聊送重命名工具)

3.labelimg标注（关注私聊送labelimg.exe工具)

操作指南：

1.点击左上角第二个选项打开目录，选择你存放图片的目录。
第三个选项改变存放目录，是存放标签的目录，你也可以点击进行选择。

2.从上图中可以看到PascalVOC这个按钮，说明当前的储存格式是VOC的xml格式，还可以点击此按钮进行储存格式的切换，支持YOLOtxt格式。

3.英文输入法下，点击快捷键w，开始创建矩形框。

4.英文输入法下，点击快捷键d，切换到下一张图片，继续标注。

5.编辑data/predefined_classes.txt文件来加载预定义的类。

6.可以在view设置每次自动保存和单类别标注。

4.数据增强扩充数据集（关注私聊送目标分割数据增强代码）

增强方式：1.裁剪；2.平移；3.亮度；4.噪声；5.旋转；6.镜像；7.cutout遮挡

（参数可改，混搭编辑，每张图片生成图片数可改）

其他增强方式如HSV色彩变化、变形缩放、背景替换等也可以适用于工业作业环境。

数据增强后记得将图片顺序化重命名！并且图片与标签名字对应！

# -*- coding=utf-8 -*-
##############################################################
# description:
#     data augmentation for obeject detection
# author:
#     pureyang 2019-08-26
# 参考：https://github.com/maozezhong/CV_ToolBox/blob/master/DataAugForObjectDetection

##############################################################

# 包括:
#     1. 裁剪(需改变bbox)
#     2. 平移(需改变bbox)
#     3. 改变亮度
#     4. 加噪声
#     5. 旋转角度(需要改变bbox)
#     6. 镜像(需要改变bbox)
#     7. cutout
#  注意:
#     random.seed(),相同的seed,产生的随机数是一样的!!


import time
import random
import copy
import cv2
import os
import math
import numpy as np
from skimage.util import random_noise
from lxml import etree, objectify
import xml.etree.ElementTree as ET
import argparse


# 显示图片
def show_pic(img, bboxes=None):
    '''
    输入:
        img:图像array
        bboxes:图像的所有boudning box list, 格式为[[x_min, y_min, x_max, y_max]....]
        names:每个box对应的名称
    '''
    for i in range(len(bboxes)):
        bbox = bboxes[i]
        x_min = bbox[0]
        y_min = bbox[1]
        x_max = bbox[2]
        y_max = bbox[3]
        cv2.rectangle(img, (int(x_min), int(y_min)), (int(x_max), int(y_max)), (0, 255, 0), 3)
    cv2.namedWindow('pic', 0)  # 1表示原图
    cv2.moveWindow('pic', 0, 0)
    cv2.resizeWindow('pic', 1200, 800)  # 可视化的图片大小
    cv2.imshow('pic', img)
    cv2.waitKey(0)
    cv2.destroyAllWindows()


# 图像均为cv2读取
class DataAugmentForObjectDetection():
    def __init__(self, rotation_rate=0.5, max_rotation_angle=5,
                 crop_rate=0.5, shift_rate=0.5, change_light_rate=0.5,
                 add_noise_rate=0.5, flip_rate=0.5,
                 cutout_rate=0.5, cut_out_length=50, cut_out_holes=1, cut_out_threshold=0.5,
                 is_addNoise=True, is_changeLight=True, is_cutout=True, is_rotate_img_bbox=True,
                 is_crop_img_bboxes=True, is_shift_pic_bboxes=True, is_filp_pic_bboxes=True):

        # 配置各个操作的属性
        self.rotation_rate = rotation_rate
        self.max_rotation_angle = max_rotation_angle
        self.crop_rate = crop_rate
        self.shift_rate = shift_rate
        self.change_light_rate = change_light_rate
        self.add_noise_rate = add_noise_rate
        self.flip_rate = flip_rate
        self.cutout_rate = cutout_rate

        self.cut_out_length = cut_out_length
        self.cut_out_holes = cut_out_holes
        self.cut_out_threshold = cut_out_threshold

        # 是否使用某种增强方式
        self.is_addNoise = is_addNoise
        self.is_changeLight = is_changeLight
        self.is_cutout = is_cutout
        self.is_rotate_img_bbox = is_rotate_img_bbox
        self.is_crop_img_bboxes = is_crop_img_bboxes
        self.is_shift_pic_bboxes = is_shift_pic_bboxes
        self.is_filp_pic_bboxes = is_filp_pic_bboxes

    # 加噪声
    def _addNoise(self, img):
        '''
        输入:
            img:图像array
        输出:
            加噪声后的图像array,由于输出的像素是在[0,1]之间,所以得乘以255
        '''
        # return cv2.GaussianBlur(img, (11, 11), 0)
        return random_noise(img, mode='gaussian', seed=int(time.time()), clip=True) * 255

    # 调整亮度
    def _changeLight(self, img):
        alpha = random.uniform(0.35, 1)
        blank = np.zeros(img.shape, img.dtype)
        return cv2.addWeighted(img, alpha, blank, 1 - alpha, 0)

    # cutout
    def _cutout(self, img, bboxes, length=100, n_holes=1, threshold=0.5):
        '''
        原版本：https://github.com/uoguelph-mlrg/Cutout/blob/master/util/cutout.py
        Randomly mask out one or more patches from an image.
        Args:
            img : a 3D numpy array,(h,w,c)
            bboxes : 框的坐标
            n_holes (int): Number of patches to cut out of each image.
            length (int): The length (in pixels) of each square patch.
        '''

        def cal_iou(boxA, boxB):
            '''
            boxA, boxB为两个框，返回iou
            boxB为bouding box
            '''
            # determine the (x, y)-coordinates of the intersection rectangle
            xA = max(boxA[0], boxB[0])
            yA = max(boxA[1], boxB[1])
            xB = min(boxA[2], boxB[2])
            yB = min(boxA[3], boxB[3])

            if xB <= xA or yB <= yA:
                return 0.0

            # compute the area of intersection rectangle
            interArea = (xB - xA + 1) * (yB - yA + 1)

            # compute the area of both the prediction and ground-truth
            # rectangles
            boxAArea = (boxA[2] - boxA[0] + 1) * (boxA[3] - boxA[1] + 1)
            boxBArea = (boxB[2] - boxB[0] + 1) * (boxB[3] - boxB[1] + 1)
            iou = interArea / float(boxBArea)
            return iou

        # 得到h和w
        if img.ndim == 3:
            h, w, c = img.shape
        else:
            _, h, w, c = img.shape
        mask = np.ones((h, w, c), np.float32)
        for n in range(n_holes):
            chongdie = True  # 看切割的区域是否与box重叠太多
            while chongdie:
                y = np.random.randint(h)
                x = np.random.randint(w)

                y1 = np.clip(y - length // 2, 0,
                             h)  # numpy.clip(a, a_min, a_max, out=None), clip这个函数将将数组中的元素限制在a_min, a_max之间，大于a_max的就使得它等于 a_max，小于a_min,的就使得它等于a_min
                y2 = np.clip(y + length // 2, 0, h)
                x1 = np.clip(x - length // 2, 0, w)
                x2 = np.clip(x + length // 2, 0, w)

                chongdie = False
                for box in bboxes:
                    if cal_iou([x1, y1, x2, y2], box) > threshold:
                        chongdie = True
                        break
            mask[y1: y2, x1: x2, :] = 0.
        img = img * mask
        return img

    # 旋转
    def _rotate_img_bbox(self, img, bboxes, angle=5, scale=1.):
        '''
        参考:https://blog.csdn.net/u014540717/article/details/53301195crop_rate
        输入:
            img:图像array,(h,w,c)
            bboxes:该图像包含的所有boundingboxs,一个list,每个元素为[x_min, y_min, x_max, y_max],要确保是数值
            angle:旋转角度
            scale:默认1
        输出:
            rot_img:旋转后的图像array
            rot_bboxes:旋转后的boundingbox坐标list
        '''
        # ---------------------- 旋转图像 ----------------------
        w = img.shape[1]
        h = img.shape[0]
        # 角度变弧度
        rangle = np.deg2rad(angle)  # angle in radians
        # now calculate new image width and height
        nw = (abs(np.sin(rangle) * h) + abs(np.cos(rangle) * w)) * scale
        nh = (abs(np.cos(rangle) * h) + abs(np.sin(rangle) * w)) * scale
        # ask OpenCV for the rotation matrix
        rot_mat = cv2.getRotationMatrix2D((nw * 0.5, nh * 0.5), angle, scale)
        # calculate the move from the old center to the new center combined
        # with the rotation
        rot_move = np.dot(rot_mat, np.array([(nw - w) * 0.5, (nh - h) * 0.5, 0]))
        # the move only affects the translation, so update the translation
        rot_mat[0, 2] += rot_move[0]
        rot_mat[1, 2] += rot_move[1]
        # 仿射变换
        rot_img = cv2.warpAffine(img, rot_mat, (int(math.ceil(nw)), int(math.ceil(nh))), flags=cv2.INTER_LANCZOS4)

        # ---------------------- 矫正bbox坐标 ----------------------
        # rot_mat是最终的旋转矩阵
        # 获取原始bbox的四个中点，然后将这四个点转换到旋转后的坐标系下
        rot_bboxes = list()
        for bbox in bboxes:
            xmin = bbox[0]
            ymin = bbox[1]
            xmax = bbox[2]
            ymax = bbox[3]
            point1 = np.dot(rot_mat, np.array([(xmin + xmax) / 2, ymin, 1]))
            point2 = np.dot(rot_mat, np.array([xmax, (ymin + ymax) / 2, 1]))
            point3 = np.dot(rot_mat, np.array([(xmin + xmax) / 2, ymax, 1]))
            point4 = np.dot(rot_mat, np.array([xmin, (ymin + ymax) / 2, 1]))
            # 合并np.array
            concat = np.vstack((point1, point2, point3, point4))
            # 改变array类型
            concat = concat.astype(np.int32)
            # 得到旋转后的坐标
            rx, ry, rw, rh = cv2.boundingRect(concat)
            rx_min = rx
            ry_min = ry
            rx_max = rx + rw
            ry_max = ry + rh
            # 加入list中
            rot_bboxes.append([rx_min, ry_min, rx_max, ry_max])

        return rot_img, rot_bboxes

    # 裁剪
    def _crop_img_bboxes(self, img, bboxes):
        '''
        裁剪后的图片要包含所有的框
        输入:
            img:图像array
            bboxes:该图像包含的所有boundingboxs,一个list,每个元素为[x_min, y_min, x_max, y_max],要确保是数值
        输出:
            crop_img:裁剪后的图像array
            crop_bboxes:裁剪后的bounding box的坐标list
        '''
        # ---------------------- 裁剪图像 ----------------------
        w = img.shape[1]
        h = img.shape[0]
        x_min = w  # 裁剪后的包含所有目标框的最小的框
        x_max = 0
        y_min = h
        y_max = 0
        for bbox in bboxes:
            x_min = min(x_min, bbox[0])
            y_min = min(y_min, bbox[1])
            x_max = max(x_max, bbox[2])
            y_max = max(y_max, bbox[3])

        d_to_left = x_min  # 包含所有目标框的最小框到左边的距离
        d_to_right = w - x_max  # 包含所有目标框的最小框到右边的距离
        d_to_top = y_min  # 包含所有目标框的最小框到顶端的距离
        d_to_bottom = h - y_max  # 包含所有目标框的最小框到底部的距离

        # 随机扩展这个最小框
        crop_x_min = int(x_min - random.uniform(0, d_to_left))
        crop_y_min = int(y_min - random.uniform(0, d_to_top))
        crop_x_max = int(x_max + random.uniform(0, d_to_right))
        crop_y_max = int(y_max + random.uniform(0, d_to_bottom))

        # 随机扩展这个最小框 , 防止别裁的太小
        # crop_x_min = int(x_min - random.uniform(d_to_left//2, d_to_left))
        # crop_y_min = int(y_min - random.uniform(d_to_top//2, d_to_top))
        # crop_x_max = int(x_max + random.uniform(d_to_right//2, d_to_right))
        # crop_y_max = int(y_max + random.uniform(d_to_bottom//2, d_to_bottom))

        # 确保不要越界
        crop_x_min = max(0, crop_x_min)
        crop_y_min = max(0, crop_y_min)
        crop_x_max = min(w, crop_x_max)
        crop_y_max = min(h, crop_y_max)

        crop_img = img[crop_y_min:crop_y_max, crop_x_min:crop_x_max]

        # ---------------------- 裁剪boundingbox ----------------------
        # 裁剪后的boundingbox坐标计算
        crop_bboxes = list()
        for bbox in bboxes:
            crop_bboxes.append([bbox[0] - crop_x_min, bbox[1] - crop_y_min, bbox[2] - crop_x_min, bbox[3] - crop_y_min])

        return crop_img, crop_bboxes

    # 平移
    def _shift_pic_bboxes(self, img, bboxes):
        '''
        参考:https://blog.csdn.net/sty945/article/details/79387054
        平移后的图片要包含所有的框
        输入:
            img:图像array
            bboxes:该图像包含的所有boundingboxs,一个list,每个元素为[x_min, y_min, x_max, y_max],要确保是数值
        输出:
            shift_img:平移后的图像array
            shift_bboxes:平移后的bounding box的坐标list
        '''
        # ---------------------- 平移图像 ----------------------
        w = img.shape[1]
        h = img.shape[0]
        x_min = w  # 裁剪后的包含所有目标框的最小的框
        x_max = 0
        y_min = h
        y_max = 0
        for bbox in bboxes:
            x_min = min(x_min, bbox[0])
            y_min = min(y_min, bbox[1])
            x_max = max(x_max, bbox[2])
            y_max = max(y_max, bbox[3])

        d_to_left = x_min  # 包含所有目标框的最大左移动距离
        d_to_right = w - x_max  # 包含所有目标框的最大右移动距离
        d_to_top = y_min  # 包含所有目标框的最大上移动距离
        d_to_bottom = h - y_max  # 包含所有目标框的最大下移动距离

        x = random.uniform(-(d_to_left - 1) / 3, (d_to_right - 1) / 3)
        y = random.uniform(-(d_to_top - 1) / 3, (d_to_bottom - 1) / 3)

        M = np.float32([[1, 0, x], [0, 1, y]])  # x为向左或右移动的像素值,正为向右负为向左; y为向上或者向下移动的像素值,正为向下负为向上
        shift_img = cv2.warpAffine(img, M, (img.shape[1], img.shape[0]))

        # ---------------------- 平移boundingbox ----------------------
        shift_bboxes = list()
        for bbox in bboxes:
            shift_bboxes.append([bbox[0] + x, bbox[1] + y, bbox[2] + x, bbox[3] + y])

        return shift_img, shift_bboxes

    # 镜像
    def _filp_pic_bboxes(self, img, bboxes):
        '''
            参考:https://blog.csdn.net/jningwei/article/details/78753607
            平移后的图片要包含所有的框
            输入:
                img:图像array
                bboxes:该图像包含的所有boundingboxs,一个list,每个元素为[x_min, y_min, x_max, y_max],要确保是数值
            输出:
                flip_img:平移后的图像array
                flip_bboxes:平移后的bounding box的坐标list
        '''
        # ---------------------- 翻转图像 ----------------------

        flip_img = copy.deepcopy(img)
        h, w, _ = img.shape

        sed = random.random()

        if 0 < sed < 0.33:  # 0.33的概率水平翻转，0.33的概率垂直翻转,0.33是对角反转
            flip_img = cv2.flip(flip_img, 0)  # _flip_x
            inver = 0
        elif 0.33 < sed < 0.66:
            flip_img = cv2.flip(flip_img, 1)  # _flip_y
            inver = 1
        else:
            flip_img = cv2.flip(flip_img, -1)  # flip_x_y
            inver = -1

        # ---------------------- 调整boundingbox ----------------------
        flip_bboxes = list()
        for box in bboxes:
            x_min = box[0]
            y_min = box[1]
            x_max = box[2]
            y_max = box[3]
            if inver == 0:
                flip_bboxes.append([x_max, h - y_min, x_min, h - y_max])
            elif inver == 1:
                flip_bboxes.append([w - x_max, y_min, w - x_min, y_max])
            elif inver == -1:
                flip_bboxes.append([w - x_min, h - y_max, w - x_max, h - y_min])

        return flip_img, flip_bboxes

    # 图像增强方法
    def dataAugment(self, img, bboxes):
        '''
        图像增强
        输入:
            img:图像array
            bboxes:该图像的所有框坐标
        输出:
            img:增强后的图像
            bboxes:增强后图片对应的box
        '''
        change_num = 0  # 改变的次数
        # print('------')
        while change_num < 1:  # 默认至少有一种数据增强生效

            if self.is_rotate_img_bbox:
                if random.random() > self.rotation_rate:  # 旋转
                    change_num += 1
                    angle = random.uniform(-self.max_rotation_angle, self.max_rotation_angle)
                    scale = random.uniform(0.7, 0.8)
                    img, bboxes = self._rotate_img_bbox(img, bboxes, angle, scale)

            if self.is_shift_pic_bboxes:
                if random.random() < self.shift_rate:  # 平移
                    change_num += 1
                    img, bboxes = self._shift_pic_bboxes(img, bboxes)

            if self.is_changeLight:
                if random.random() > self.change_light_rate:  # 改变亮度
                    change_num += 1
                    img = self._changeLight(img)

            if self.is_addNoise:
                if random.random() < self.add_noise_rate:  # 加噪声
                    change_num += 1
                    img = self._addNoise(img)
            if self.is_cutout:
                if random.random() < self.cutout_rate:  # cutout
                    change_num += 1
                    img = self._cutout(img, bboxes, length=self.cut_out_length, n_holes=self.cut_out_holes,
                                       threshold=self.cut_out_threshold)
            if self.is_filp_pic_bboxes:
                if random.random() < self.flip_rate:  # 翻转
                    change_num += 1
                    img, bboxes = self._filp_pic_bboxes(img, bboxes)

        return img, bboxes


# xml解析工具
class ToolHelper():
    # 从xml文件中提取bounding box信息, 格式为[[x_min, y_min, x_max, y_max, name]]
    def parse_xml(self, path):
        '''
        输入：
            xml_path: xml的文件路径
        输出：
            从xml文件中提取bounding box信息, 格式为[[x_min, y_min, x_max, y_max, name]]
        '''
        tree = ET.parse(path)
        root = tree.getroot()
        objs = root.findall('object')
        coords = list()
        for ix, obj in enumerate(objs):
            name = obj.find('name').text
            box = obj.find('bndbox')
            x_min = int(box[0].text)
            y_min = int(box[1].text)
            x_max = int(box[2].text)
            y_max = int(box[3].text)
            coords.append([x_min, y_min, x_max, y_max, name])
        return coords

    # 保存图片结果
    def save_img(self, file_name, save_folder, img):
        cv2.imwrite(os.path.join(save_folder, file_name), img)

    # 保持xml结果
    def save_xml(self, file_name, save_folder, img_info, height, width, channel, bboxs_info):
        '''
        :param file_name:文件名
        :param save_folder:#保存的xml文件的结果
        :param height:图片的信息
        :param width:图片的宽度
        :param channel:通道
        :return:
        '''
        folder_name, img_name = img_info  # 得到图片的信息

        E = objectify.ElementMaker(annotate=False)

        anno_tree = E.annotation(
            E.folder(folder_name),
            E.filename(img_name),
            E.path(os.path.join(folder_name, img_name)),
            E.source(
                E.database('Unknown'),
            ),
            E.size(
                E.width(width),
                E.height(height),
                E.depth(channel)
            ),
            E.segmented(0),
        )

        labels, bboxs = bboxs_info  # 得到边框和标签信息
        for label, box in zip(labels, bboxs):
            anno_tree.append(
                E.object(
                    E.name(label),
                    E.pose('Unspecified'),
                    E.truncated('0'),
                    E.difficult('0'),
                    E.bndbox(
                        E.xmin(box[0]),
                        E.ymin(box[1]),
                        E.xmax(box[2]),
                        E.ymax(box[3])
                    )
                ))

        etree.ElementTree(anno_tree).write(os.path.join(save_folder, file_name), pretty_print=True)


if __name__ == '__main__':

    need_aug_num = 10  # 每张图片需要增强的次数

    is_endwidth_dot = True  # 文件是否以.jpg或者png结尾

    dataAug = DataAugmentForObjectDetection()  # 数据增强工具类

    toolhelper = ToolHelper()  # 工具

    # 获取相关参数
    parser = argparse.ArgumentParser()
    parser.add_argument('--source_img_path', type=str, default='data/Images')
    parser.add_argument('--source_xml_path', type=str, default='data/Annotations')
    parser.add_argument('--save_img_path', type=str, default='data/Images2')
    parser.add_argument('--save_xml_path', type=str, default='data/Annotations2')
    args = parser.parse_args()
    source_img_path = args.source_img_path  # 图片原始位置
    source_xml_path = args.source_xml_path  # xml的原始位置

    save_img_path = args.save_img_path  # 图片增强结果保存文件
    save_xml_path = args.save_xml_path  # xml增强结果保存文件

    # 如果保存文件夹不存在就创建
    if not os.path.exists(save_img_path):
        os.mkdir(save_img_path)

    if not os.path.exists(save_xml_path):
        os.mkdir(save_xml_path)

    for parent, _, files in os.walk(source_img_path):
        files.sort()
        for file in files:
            cnt = 0
            pic_path = os.path.join(parent, file)
            xml_path = os.path.join(source_xml_path, file[:-4] + '.xml')
            values = toolhelper.parse_xml(xml_path)  # 解析得到box信息，格式为[[x_min,y_min,x_max,y_max,name]]
            coords = [v[:4] for v in values]  # 得到框
            labels = [v[-1] for v in values]  # 对象的标签

            # 如果图片是有后缀的
            if is_endwidth_dot:
                # 找到文件的最后名字
                dot_index = file.rfind('.')
                _file_prefix = file[:dot_index]  # 文件名的前缀
                _file_suffix = file[dot_index:]  # 文件名的后缀
            img = cv2.imread(pic_path)

            # show_pic(img, coords)  # 显示原图
            while cnt < need_aug_num:  # 继续增强
                auged_img, auged_bboxes = dataAug.dataAugment(img, coords)
                auged_bboxes_int = np.array(auged_bboxes).astype(np.int32)
                height, width, channel = auged_img.shape  # 得到图片的属性
                img_name = '{}_{}{}'.format(_file_prefix, cnt + 1, _file_suffix)  # 图片保存的信息
                toolhelper.save_img(img_name, save_img_path,
                                    auged_img)  # 保存增强图片

                toolhelper.save_xml('{}_{}.xml'.format(_file_prefix, cnt + 1),
                                    save_xml_path, (save_img_path, img_name), height, width, channel,
                                    (labels, auged_bboxes_int))  # 保存xml文件
                # show_pic(auged_img, auged_bboxes)  # 强化后的图
                print(img_name)
                cnt += 1  # 继续增强下一张

5.YOLOv5数据集制作及配置

5.1训练集、验证集、测试集的划分，

通过脚本生成，创建split_train_val.py文件，代码内容如下：

# coding:utf-8

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='ImageSets/Main', type=str, help='output txt label path')
opt = parser.parse_args()

trainval_percent = 1.0
train_percent = 0.9
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_index = range(num)
tv = int(num * trainval_percent)
tr = int(tv * train_percent)
trainval = random.sample(list_index, tv)
train = random.sample(trainval, tr)

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

for i in list_index:
    name = total_xml[i][:-4] + '\n'
    if i in trainval:
        file_trainval.write(name)
        if i in train:
            file_train.write(name)
        else:
            file_val.write(name)
    else:
        file_test.write(name)

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

5.2把数据集格式转换成yolo_txt格式

# -*- coding: utf-8 -*-
import xml.etree.ElementTree as ET
import os
from os import getcwd

sets = ['train', 'val', 'test']
classes = ["yes", "back"]  # 改成自己的类别
abs_path = os.getcwd()
print(abs_path)


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('D:/G/YOLOdata2/photos/Annotations/%s.xml' % (image_id), encoding='UTF-8')
    out_file = open('D:/G/YOLOdata2/photos/labels/%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')


wd = getcwd()
for image_set in sets:
    if not os.path.exists('D:/G/YOLOdata2/photos/labels/'):
        os.makedirs('D:/G/YOLOdata2/photos/labels/')
    image_ids = open('D:/G/YOLOdata2/photos/ImageSets/Main/%s.txt' % (image_set)).read().strip().split()

    if not os.path.exists('D:/G/YOLOdata2/photos/dataSet_path/'):
        os.makedirs('D:/G/YOLOdata2/photos/dataSet_path/')

    list_file = open('dataSet_path/%s.txt' % (image_set), 'w')
    # 这行路径不需更改，这是相对路径
    for image_id in image_ids:
        list_file.write('D:/G/YOLOdata2/photos/images/%s.jpg\n' % (image_id))
        convert_annotation(image_id)
    list_file.close()

5.3 在 yolov5 目录下的 data 文件夹下新建一个 myvoc.yaml文件

train: /mnt/paper_data/dataSet_path/train.txt
val: /mnt/paper_data/dataSet_path/val.txt

# number of classes
nc: 2

# class names
names: ["yes", "back"]

5.4聚类获得先验框

在photos目录下创建程序两个程序 kmeans.py ，

以及 clauculate_anchors.py，运行 clauculate_anchors.py 即可。

5.4.1kmeans代码：

import numpy as np

def iou(box, clusters):
    """
    Calculates the Intersection over Union (IoU) between a box and k clusters.
    :param box: tuple or array, shifted to the origin (i. e. width and height)
    :param clusters: numpy array of shape (k, 2) where k is the number of clusters
    :return: numpy array of shape (k, 0) where k is the number of clusters
    """
    x = np.minimum(clusters[:, 0], box[0])
    y = np.minimum(clusters[:, 1], box[1])
    if np.count_nonzero(x == 0) > 0 or np.count_nonzero(y == 0) > 0:
        raise ValueError("Box has no area")    # 如果报这个错，可以把这行改成pass即可

    intersection = x * y
    box_area = box[0] * box[1]
    cluster_area = clusters[:, 0] * clusters[:, 1]

    iou_ = intersection / (box_area + cluster_area - intersection)

    return iou_

def avg_iou(boxes, clusters):
    """
    Calculates the average Intersection over Union (IoU) between a numpy array of boxes and k clusters.
    :param boxes: numpy array of shape (r, 2), where r is the number of rows
    :param clusters: numpy array of shape (k, 2) where k is the number of clusters
    :return: average IoU as a single float
    """
    return np.mean([np.max(iou(boxes[i], clusters)) for i in range(boxes.shape[0])])

def translate_boxes(boxes):
    """
    Translates all the boxes to the origin.
    :param boxes: numpy array of shape (r, 4)
    :return: numpy array of shape (r, 2)
    """
    new_boxes = boxes.copy()
    for row in range(new_boxes.shape[0]):
        new_boxes[row][2] = np.abs(new_boxes[row][2] - new_boxes[row][0])
        new_boxes[row][3] = np.abs(new_boxes[row][3] - new_boxes[row][1])
    return np.delete(new_boxes, [0, 1], axis=1)


def kmeans(boxes, k, dist=np.median):
    """
    Calculates k-means clustering with the Intersection over Union (IoU) metric.
    :param boxes: numpy array of shape (r, 2), where r is the number of rows
    :param k: number of clusters
    :param dist: distance function
    :return: numpy array of shape (k, 2)
    """
    rows = boxes.shape[0]

    distances = np.empty((rows, k))
    last_clusters = np.zeros((rows,))

    np.random.seed()

    # the Forgy method will fail if the whole array contains the same rows
    clusters = boxes[np.random.choice(rows, k, replace=False)]

    while True:
        for row in range(rows):
            distances[row] = 1 - iou(boxes[row], clusters)

        nearest_clusters = np.argmin(distances, axis=1)

        if (last_clusters == nearest_clusters).all():
            break

        for cluster in range(k):
            clusters[cluster] = dist(boxes[nearest_clusters == cluster], axis=0)

        last_clusters = nearest_clusters

    return clusters

if __name__ == '__main__':
    a = np.array([[1, 2, 3, 4], [5, 7, 6, 8]])
    print(translate_boxes(a))

5.4.2 clauculate_anchors.py代码

# -*- coding: utf-8 -*-
# 根据标签文件求先验框

import os
import numpy as np
import xml.etree.cElementTree as et
from kmeans import kmeans, avg_iou

FILE_ROOT = "D:/G/YOLOdata2/photos/"     # 根路径
ANNOTATION_ROOT = "Annotations"   # 数据集标签文件夹路径
ANNOTATION_PATH = FILE_ROOT + ANNOTATION_ROOT

ANCHORS_TXT_PATH = "D:/G/YOLOdata2/photos/anchors.txt"   #anchors文件保存位置

CLUSTERS = 9
CLASS_NAMES = ['yes', 'back']   #类别名称

def load_data(anno_dir, class_names):
    xml_names = os.listdir(anno_dir)
    boxes = []
    for xml_name in xml_names:
        xml_pth = os.path.join(anno_dir, xml_name)
        tree = et.parse(xml_pth)

        width = float(tree.findtext("./size/width"))
        height = float(tree.findtext("./size/height"))

        for obj in tree.findall("./object"):
            cls_name = obj.findtext("name")
            if cls_name in class_names:
                xmin = float(obj.findtext("bndbox/xmin")) / width
                ymin = float(obj.findtext("bndbox/ymin")) / height
                xmax = float(obj.findtext("bndbox/xmax")) / width
                ymax = float(obj.findtext("bndbox/ymax")) / height

                box = [xmax - xmin, ymax - ymin]
                boxes.append(box)
            else:
                continue
    return np.array(boxes)

if __name__ == '__main__':

    anchors_txt = open(ANCHORS_TXT_PATH, "w")

    train_boxes = load_data(ANNOTATION_PATH, CLASS_NAMES)
    count = 1
    best_accuracy = 0
    best_anchors = []
    best_ratios = []

    for i in range(10):      ##### 可以修改，不要太大，否则时间很长
        anchors_tmp = []
        clusters = kmeans(train_boxes, k=CLUSTERS)
        idx = clusters[:, 0].argsort()
        clusters = clusters[idx]
        # print(clusters)

        for j in range(CLUSTERS):
            anchor = [round(clusters[j][0] * 640, 2), round(clusters[j][1] * 640, 2)]
            anchors_tmp.append(anchor)
            print(f"Anchors:{anchor}")

        temp_accuracy = avg_iou(train_boxes, clusters) * 100
        print("Train_Accuracy:{:.2f}%".format(temp_accuracy))

        ratios = np.around(clusters[:, 0] / clusters[:, 1], decimals=2).tolist()
        ratios.sort()
        print("Ratios:{}".format(ratios))
        print(20 * "*" + " {} ".format(count) + 20 * "*")

        count += 1

        if temp_accuracy > best_accuracy:
            best_accuracy = temp_accuracy
            best_anchors = anchors_tmp
            best_ratios = ratios

    anchors_txt.write("Best Accuracy = " + str(round(best_accuracy, 2)) + '%' + "\r\n")
    anchors_txt.write("Best Anchors = " + str(best_anchors) + "\r\n")
    anchors_txt.write("Best Ratios = " + str(best_ratios))
    anchors_txt.close()

5.5修改model/yolov5s.yalm文件修改类别nc和anchor尺寸

6.训练train.py

6.1修改train.py参数

epochs：指的就是训练过程中整个数据集将被迭代多少次,显卡不行你就调小点。
batch-size：一次看完多少张图片才进行权重更新，梯度下降的mini-batch,显卡不行你就调小点。
cfg：存储模型结构的配置文件
data：存储训练、测试数据的文件
img-size：输入图片宽高,显卡不行你就调小点。
rect：进行矩形训练
resume：恢复最近保存的模型开始训练
nosave：仅保存最终checkpoint
notest：仅测试最后的epoch
evolve：进化超参数
bucket：gsutil bucket
cache-images：缓存图像以加快训练速度
weights：权重文件路径
name：重命名results.txt to results_name.txt
device：cuda device, i.e. 0 or 0,1,2,3 or cpu
adam：使用adam优化
multi-scale：多尺度训练，img-size +/- 50%
single-cls：单类别的训练集
————————————————

训练命令：

python train.py --img 640 --batch 32 --epoch 300 --data data/myvoc.yaml --cfg models/yolov5s.yaml --weights yolov5s.pt --device 0

6.2修改val.py参数

6.3修改dectet.py参数

7.Result结果分析

具体可以看这篇文章写的挺全的：

【目标检测算法】YOLO-V5训练结果的分析与评价_yolov5训练结果分析_三木今天学习了嘛的博客-CSDN博客