yolov5训练步骤及安全帽检测

环境部署问题、训练后无法识别问题都有介绍注意事项

一、说明

• 系统uname -ar：ubuntu18.0.4(Linux ubuntu 5.4.0-122-generic) 64bit
• 显卡lspci：GeForce GT 1030
• nvidia版本：NVIDIA-Linux-x86_64-470.129.06.run (该网址搜索下载：https://www.nvidia.cn/geforce/drivers)
• cuda版本： cuda_10.2.89_440.33.01_linux.run (历史版本：https://developer.nvidia.com/cuda-toolkit-archive)
• yolov5: 代码是tag v6.1

二、PC机nvidia显卡(没有忽略用CPU方式)

1. 禁用 nouveau驱动

   lsmod | grep nouveau
   		禁用:
   		sudo vim /etc/modprobe.d/blacklist.conf
   			blacklist nouveau
   			options nouveau modeset=0
   		
   		sudo update-initramfs -u
   

2. nvidia驱动安装

   sudo apt install dkms build-essential linux-headers-generic
   	sudo apt-get install -y libc6-i386 lib32stdc++6 lib32gcc1 lib32ncurses5 lib32z1
   	sudo ./NVIDIA-Linux-x86_64-470.129.06.run --dkms --no-opengl-files
   	
   	一定操作，否则会失败
   		在BIOS界面,禁用secure boot(安全模式)
   	
   	(安装失败重装)
   	sudo nvidia-uninstall
   	sudo apt-get remove --purge nvidia*
   

3. 安装cuda10.2

   A、sudo service lightdm stop
   
   B、如果已安装nvidia驱动，安装时把驱动取消,建议先安装驱动，并且驱动版本大于cuda后缀的440.33.01，否则cuda安装失败)
   		wget https://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda_10.2.89_440.33.01_linux.run
   		sudo sh cuda_10.2.89_440.33.01_linux.run --no-opengl-libs (如果开始安装了nvidia-390.151版本驱动会有问题)
   C、vi ~/.bashrc
   		export PATH="/usr/local/cuda-10.2/bin:$PATH" 
   		export LD_LIBRARY_PATH="/usr/local/cuda-10.2/lib64:$LD_LIBRARY_PATH"
   	
   D、sudo service lightdm start
   

4. nvidia-smi：失败

   ls /usr/src | grep nvidia  #查看自己安装的nvidia版本，我的是470.129.06
   sudo apt install dkms
   sudo dkms install -m nvidia -v 470.129.06 
   

三、python升级

1. python3.6–>pyhton3.9 (系统自带3.6)
2. 安装依赖

   sudo apt-get install libffi-dev zlib1g-dev libbz2-dev libssl-dev liblzma-dev
   

3. 下载python3.9

   wget https://www.python.org/ftp/python/3.9.0/Python-3.9.0.tgz
   

4. 编译

   ./configure --with-ssl --enable-optimizations (--with-ssl 参数要加上否则使用中会出错)
   	make
   	sudo make install
   

5. 设置软连接

   sudo ln -s /usr/local/bin/python3 /usr/bin/python39 #不要修改系统python3软连接，否则一些命令无法使用
   sudo ln -s /usr/local/bin/pip3 /usr/bin/pip3
   

四、pytorch安装

1. 官网

   https://pytorch.org/get-started/locally/
   

2. CPU

   pip install torch==1.9.0+cpu torchvision==0.10.0+cpu torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html
   如果失败进入https://download.pytorch.org/whl/torch_stable.html 下载对应版本
   

3. GPU

   pip install torch==1.9.0+cu102 torchvision==0.10.0+cu102 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html
   	
   pip3 install torch-1.9.0+cu102-cp39-cp39-linux_x86_64.whl
   pip3 install torchvision-0.10.0+cu102-cp39-cp39-linux_x86_64.whl
   pip3 install torchaudio-0.9.0-cp39-cp39-linux_x86_64.whl
   

五、pip 依赖

absl-py==0.12.0
altgraph==0.17
backcall==0.2.0
backports.lzma==0.0.14
cachetools==4.2.1
certifi==2020.12.5
chardet==4.0.0
charset-normalizer==2.1.0
click==8.1.3
cycler==0.10.0
decorator==5.1.1
docker-pycreds==0.4.0
future==0.18.2
gitdb==4.0.9
GitPython==3.1.27
google-auth==1.28.1
google-auth-oauthlib==0.4.4
GPUtil==1.4.0
grpcio==1.37.0
idna==2.10
importlib-metadata==4.12.0
ipython==7.34.0
jedi==0.18.1
kiwisolver==1.3.1
lxml==4.9.1
Markdown==3.4.1
matplotlib==3.3.4
matplotlib-inline==0.1.3
numpy==1.21.6
oauthlib==3.2.0
opencv-python==4.5.1.48
pandas==1.3.5
parso==0.8.3
pathtools==0.1.2
pexpect==4.8.0
pickleshare==0.7.5
Pillow==9.2.0
promise==2.3
prompt-toolkit==3.0.30
protobuf==3.15.8
psutil==5.9.1
ptyprocess==0.7.0
pyasn1==0.4.8
pyasn1-modules==0.2.8
Pygments==2.12.0
pyparsing==3.0.9
PyQt5==5.15.4
pyqt5-plugins==5.15.4.2.2
PyQt5-Qt5==5.15.2
PyQt5-sip==12.11.0
pyqt5-tools==5.15.4.3.2
PyQtChart==5.15.4
PyQtChart-Qt5==5.15.2
python-dateutil==2.8.2
python-dotenv==0.20.0
pytz==2022.1
PyYAML==6.0
qt5-applications==5.15.2.2.2
qt5-tools==5.15.2.1.2
requests==2.25.1
requests-oauthlib==1.3.1
rsa==4.9
scipy==1.6.1
seaborn==0.11.2
sentry-sdk==1.8.0
setproctitle==1.2.3
shortuuid==1.0.9
six==1.16.0
smmap==5.0.0
tensorboard==2.4.1
tensorboard-plugin-wit==1.8.1
tornado==6.1
tqdm==4.64.0
traitlets==5.3.0
typing-extensions==4.3.0
urllib3==1.26.5
wandb==0.12.21
wcwidth==0.2.5
Werkzeug==1.0.1
wincertstore==0.2
zipp==3.8.1

六、准备数据集

1. YOLOv5代码

   git clone https://github.com/ultralytics/yolov5
   

2. 图片资源，用飞桨安全帽资源，已经标注好了

   HelmetDetection包括images(原始图片)和annotations(标注信息xml) 
   下载地址：https://aistudio.baidu.com/aistudio/datasetdetail/50329
   

3. yolov5中创建目录(资源转成VOC格式)

   helmet_source
   		Annotations  #标注信息xml
   		dataSet_path #
   		images       #原始图片
   		ImageSets    #数据集分类txt文件(自写make_voc_txt.py脚本生成)
   		labels       #voc格式的标签文件(自写make_voc_label.py脚本生成)
   

4. 注意

   1. 按照上述目录结构训练结果可以检测出图片
   2. 在data中创建的目录训练cls一直是0,检测图片也不识别
   3. 尝试换环境版本,调参数都不行
   4. 最后觉的可能哪里路径有问题
   

七、训练

1. 将coco.yaml复制一份helmet.yaml修改如下：

   train: helmet_source/dataSet_path/train.txt  # train images (relative to 'path') 118287 images
   val: helmet_source/dataSet_path/val.txt  # val images (relative to 'path') 5000 images
   # test: helmet_source/dataSet_path/test.txt  # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794
   
   # Classes
   nc: 2  # number of classes
   names: ['helmet', 'head']  # class names
   

2. 将models/yolov5s.yaml修改

   nc: 2  # number of classes  改为自己的类别个数
   

3. 训练指令(我的测试资源和脚本以及官网权重在文中后面有写)

   python39 train.py --img 416 --batch 4 --epochs 100 --data data/helmet.yaml --cfg models/yolov5s.yaml --weights weights/yolov5s.pt --device 0 #--device cpu
   

4. 说明

   输出：runs/train/exp/weights/best.pt 和 last.pt
   说明：训练100次,效果不太好,500次会好点,当然越多越好
   YOLOv5 训练 ( train.py )、验证 ( val.py )、推理 ( detect.py ) 和导出 ( export.py ) 的正确操作
   

5.训练结果图

八、测试

1. 检测指令

   python39 detect.py --data data/helmet.yaml --weights runs/train/exp/weights/best.pt --source helmet_test.png #--conf-thres 0.1 --iou-thres 0.9
   

2. 原始图片和结果
   
   

九、资源及脚本

1. 测试资源下载点击这里下载

2. 分类脚本 make_voc_txt.py

   import os
   import random
   trainval_percent = 0.1
   train_percent = 0.9
   root_path = 'helmet_source'
   xmlfilepath = '%s/Annotations' % root_path
   txtsavepath = '%s/ImageSets' % root_path
   total_xml = os.listdir(xmlfilepath)
   num = len(total_xml)
   list = range(num)
   tv = int(num * trainval_percent)
   tr = int(tv * train_percent)
   trainval = random.sample(list, tv)
   train = random.sample(trainval, tr)
   ftrainval = open('%s/trainval.txt' % txtsavepath, 'w')
   ftest = open('%s/test.txt' % txtsavepath, 'w')
   ftrain = open('%s/train.txt' % txtsavepath, 'w')
   fval = open('%s/val.txt' % txtsavepath, 'w')
   for i in list:
       name = total_xml[i][:-4] + '\n'
       if i in trainval:
           ftrainval.write(name)
           if i in train:
               ftest.write(name)
           else:
               fval.write(name)
       else:
           ftrain.write(name)
   ftrainval.close()
   ftrain.close()
   fval.close()
   ftest.close()
   
   

3. 生产yolo需要的标注数据格式make_voc_label.py（同时可以产生标注图片）

   import xml.etree.ElementTree as ET
   import pickle
   import os, cv2
   from os import listdir, getcwd
   from os.path import join
   from tqdm import tqdm
   
   sets = ['train', 'test','val']
   classes = ['helmet', 'head']
   colors = {'helmet': (60, 60, 250), 'head': (250, 60, 60)}
   
   root_path = "helmet_source"
   dataSet_path = "%s/dataSet_path" % root_path
   image_path = "%s/images" % root_path
   Annotations_path = "%s/Annotations" % root_path
   ImageSets_path = "%s/ImageSets" % root_path
   labels_path = "%s/labels" % root_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
       # x_center = (box[0]+box[1])/2.0
       # y_center = (box[2]+box[3])/2.0
       # x = x_center / size[0]
       # y = y_center / size[1]
       
       # w = (box[1] - box[0]) / size[0]
       # h = (box[3] - box[2]) / size[1]
       return (x, y, w, h)
   def convert_annotation(image_id):
       in_file = open('%s/%s.xml' % (Annotations_path, image_id))
       out_file = open('%s/%s.txt' % (labels_path, image_id), 'w')
       im = cv2.imread('%s/%s.png' % (image_path, image_id))
       print('%s/%s.png' % (image_path, image_id))
   	
       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')
           xmin = xmlbox.find('xmin').text
           xmax = xmlbox.find('xmax').text
           ymin = xmlbox.find('ymin').text
           ymax = xmlbox.find('ymax').text
           b = (float(xmin), float(xmax), float(ymin), float(ymax))
           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')
   		
           print(cls, colors[cls], xmin, xmax, ymin, ymax)
           cv2.rectangle(im, (int(xmin), int(ymin)), (int(xmax), int(ymax)), colors[cls])
           cv2.putText(im, cls, (int(xmin), int(ymin) - 3), cv2.FONT_HERSHEY_SIMPLEX, 0.5, colors[cls])
       # cv2.imshow('result', im)
       # cv2.waitKey(0)
       cv2.imwrite('%s/%s_tag.png' % (image_path, image_id), im)
   		
   wd = getcwd()
   print(wd)
   for image_set in sets:
       if not os.path.exists(labels_path):
           os.makedirs(labels_path)
       image_ids = open('%s/%s.txt' % (ImageSets_path, image_set)).read().strip().split()
       list_file = open('%s/%s.txt' % (dataSet_path, image_set), 'w')
       for image_id in image_ids:
           # print(image_id)
           list_file.write('%s/%s.png\n' % (image_path, image_id))
           convert_annotation(image_id)
       list_file.close()
   

4. 文件重命名分序make_voc_file.py

   import os
   path = "./image"
   filelist = os.listdir(path) #该文件夹下所有的文件（包括文件夹）
   count=0 #从零开始
   for file in filelist:
       print(file)
   for file in filelist:   #遍历所有文件
       Olddir=os.path.join(path,file)   #原来的文件路径
       if os.path.isdir(Olddir):   #如果是文件夹则跳过
           continue
       filename=os.path.splitext(file)[0]   #文件名
       filetype=os.path.splitext(file)[1]   #文件扩展名
       Newdir=os.path.join(path,str(count).zfill(6)+filetype)  #用字符串函数zfill 以0补全所需位数
       os.rename(Olddir,Newdir)#重命名
       count+=1
   

5.下载官方权重(里面有download_weights.sh脚本)

from utils.downloads import attempt_download

models = ['n', 's', 'm', 'l', 'x']
models.extend([x + '6' for x in models])  # add P6 models

for x in models:
	attempt_download(f'yolov5{x}.pt')

附：

1、pip3 install backports.lzma (3.9忽略)
	  sudo vi /usr/local/lib/python3.7/lzma.py
		from _lzma import *
		from _lzma import _encode_filter_properties, _decode_filter_properties
		修改：
		try:
			from _lzma import *
			from _lzma import _encode_filter_properties, _decode_filter_properties
		except ImportError:
			from backports.lzma import *
			from backports.lzma import _encode_filter_properties, _decode_filter_properties
	
2、labelImg使用
		git clone https://github.com/tzutalin/labelImg
		pip3 install lxml
		pyrcc5 -o resources.py resources.qrc , 将Qt文件格式（.qrc）转为Python（.py）格式，将生成的resources.py拷贝到同级的libs目录下
		sudo apt-get install libxcb-xinerama0 (解决 qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "" even though it was found)
		
3、pip3 运行出错：subprocess.CalledProcessError: Command '('lsb_release', '-a')' returned non-zero exit status 1  
		sudo cp /usr/lib/python3/dist-packages/lsb_release.py /usr/local/lib/python3.7  (3.9忽略)