自动驾驶图片验证

任务一

1.进入yolo的文件夹下，终端运行 pip install -r requirements.txt

2.数据标注：将官方提供的未打标图片.jpg,用labelImg进行标注，保存为.xml格式。（记得保存在指定的文件夹，最后提交官方）

3.将.xml数据上传云端，与云端数据合并（如果觉得自己打标的质量一般，可以跳过这一步）

4.将官方提供的.xml转换为yolo格式.txt(用下面这个脚本)

In [ ]:

#xml2yolo 将xml格式转换为yolo可用的txt
#!/usr/bin/env python
# -*- coding: utf8 -*-
import os
import sys
from xml.etree import ElementTree
from xml.etree.ElementTree import Element, SubElement
from lxml import etree
import codecs
import cv2
from glob import glob

XML_EXT = '.xml'
ENCODE_METHOD = 'utf-8'

class PascalVocReader:
    def __init__(self, filepath):
        # shapes type:
        # [labbel, [(x1,y1), (x2,y2), (x3,y3), (x4,y4)], color, color, difficult]
        self.shapes = []
        self.filepath = filepath
        self.verified = False
        try:
            self.parseXML()
        except:
            pass

    def getShapes(self):
        return self.shapes

    def addShape(self, label, bndbox, filename, difficult):
        xmin = int(bndbox.find('xmin').text)
        ymin = int(bndbox.find('ymin').text)
        xmax = int(bndbox.find('xmax').text)
        ymax = int(bndbox.find('ymax').text)
        points = [(xmin, ymin), (xmax, ymin), (xmax, ymax), (xmin, ymax)]
        self.shapes.append((label, points, filename, difficult))

    def parseXML(self):
        assert self.filepath.endswith(XML_EXT), "Unsupport file format"
        parser = etree.XMLParser(encoding=ENCODE_METHOD)
        xmltree = ElementTree.parse(self.filepath, parser=parser).getroot()
        filename = xmltree.find('filename').text
        path = xmltree.find('path').text
        try:
            verified = xmltree.attrib['verified']
            if verified == 'yes':
                self.verified = True
        except KeyError:
            self.verified = False

        for object_iter in xmltree.findall('object'):
            bndbox = object_iter.find("bndbox")
            label = object_iter.find('name').text
            # Add chris

            difficult = False
            if object_iter.find('difficult') is not None:
                difficult = bool(int(object_iter.find('difficult').text))
            self.addShape(label, bndbox, path, difficult)
        return True


classes = dict()
num_classes = 0

try:
    input = raw_input
except NameError:
    pass

dirpath = "../datasets/coco128/images/image_new/train_0_10_xml/" #xml路径
classes_txt = '../datasets/coco128/images/image_new/classes_txt/'  #classes.txt路径
ext = '.jpg'  #input("Image file extension: ")  '.jpg' #不用改
dirpath_img = '../datasets/coco128/images/image_new/all_img/' #xml对应的图像路径
dirpat_txt = '../datasets/labels/train/' #txt存放路径


if os.path.isfile(classes_txt):
    with open(classes_txt, "r") as f:
        class_list = f.read().strip().split()
        classes = {k : v for (v, k) in enumerate(class_list)}

filePaths = glob(dirpath_img + "/*" + ext)  #图像路径+文件
xmlNames =  os.listdir(dirpath)

# filePaths = glob(dirpath_img + "/*.xml")

#xmlName = xmlNames[0]
for xmlName in xmlNames:
    # filePath = os.path.splitext(filePath)[0]
    # fileName = os.path.splitext(filePath)[-2]
    imgFile = dirpath_img + '/' + xmlName.split(".")[0] + ".jpg"
    txtFile = dirpat_txt + '/' + xmlName.split(".")[0] + ".txt"
    xmlFile = dirpath + '/' + xmlName.split(".")[0] + ".xml"
    with open(txtFile, "w") as f:
        if not os.path.isfile(xmlFile):
            continue

        tVocParseReader = PascalVocReader(xmlFile)
        shapes = tVocParseReader.getShapes()

        for shape in shapes:
            class_name = shape[0]
            box = shape[1]
            filename = imgFile  #dirpath_img +  ext   #filePath + ext

            if class_name not in classes.keys():
                classes[class_name] = num_classes
                num_classes += 1
            class_idx = classes[class_name]
            print(filename)
            (height, width, _) = cv2.imread(filename).shape
            cv2.imread(filename)

            coord_min = box[0]
            coord_max = box[2]

            xcen = float((coord_min[0] + coord_max[0])) / 2 / width
            ycen = float((coord_min[1] + coord_max[1])) / 2 / height
            w = float((coord_max[0] - coord_min[0])) / width
            h = float((coord_max[1] - coord_min[1])) / height

            f.write("%d %.06f %.06f %.06f %.06f\n" % (class_idx, xcen, ycen, w, h))
            print(class_idx, xcen, ycen, w, h)

划分数据集：训练集和验证集

In [ ]:

#train_test_split 划分训练集验证集
import glob
import random
import os
filelist  = glob.glob('test_split/test_label/*.txt') #读取txt文件的路径
img_path = "test_split/train" #读取jpg文件的路径
test = random.sample(filelist, int(len(filelist)*0.2)) #划分比例
output_img_path = 'test_split/images/val' #验证集jpg输出路径
output_label_path = 'test_split/labels/val' #验证集labels输出路径
if not os.path.exists(output_img_path):
    os.makedirs(output_img_path)

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

for file in test:
    filename = os.path.basename(file)
    img_filename = os.path.splitext(filename)[0] + '.jpg'
    
    txtpath = file
    impath = os.path.join(img_path, img_filename)
    
    out_text = os.path.join(output_label_path, os.path.basename(txtpath))
    out_image = os.path.join(output_img_path, os.path.basename(impath))
    print(txtpath,impath,out_text,out_image)
    os.system('mv ' + txtpath + ' ' + out_text)
    os.system('mv ' + impath + ' ' + out_image)

按照官方要求存放数据

--------- images \n

-----train

-----val

---------labels \n

-----train

-----val

进入 coco128.yaml 文件 修改文件路径、classes

进入train.py，找到def oarse_opt，修改:

(1)--weights yolov5l #可选择 yolov5s/yolov5m/yolov5l

(2)--data coco128.yaml #上面已经修改过

去终端 运行train.py

训练结束后，打开detect.py,修改：

（1）--weight参数，进入runs/train/exp/weights,用weight/best.pt替换

（2）--source参数，替换为测试集"images/test"路径

（3）--data参数，如果默认为coco128.yaml则无需修改

（4）修改txt输出的格式，与官方示例保持一致

In [1]:

save_txt_add_labels yolo的输出加标签
if save_txt:  # Write to file
   xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist()  # normalized xywh
                        
   categories = ['Pedestrian', 'Cyclist', 'car', 'Truck', 'Tram' , 'Tricycle']
   labels = categories[int(cls)]
                        
   line = (cls, *xywh, conf) if save_conf else (cls, *xywh)  # label format
   with open(f"{txt_path}.txt", "a") as f:
        f.write(("%g " * len(line)).rstrip() % line + labels + "\n")

  Cell In[1], line 1
    save_txt_add_labels yolo的输出加标签
                        ^
SyntaxError: invalid syntax

进入终端，运行python detect.py --save-txt

进入run/detect/exp/labels将.txt保存到官方指定文件夹中

In [ ]:

##任务二
1.数据标注：用官方给的.jpg数据导入labelme打标，输出.json格式
2.用批处理脚本将.json转换为datasets
"yourDir=./"  #.json文件路径
"outputDir=./" #输出文件路径

In [ ]:

#labelme_json_to_dataset_in_batch 任务二图像分割批量json转dataset
setlocal
set "yourDir=./"
set "yourExt=*.json"
set "outputDir=./"
pushd %yourDir%
for %%a in (%yourExt%) do (
    labelme_json_to_dataset %%a -o "%outputDir%%%~na"
)
popd
endlocal

3.输出官方指定示例.png格式文件，存放在本地桌面

In [ ]:

linkcode

# dataset2mask
import os  
import shutil  

def copy_and_rename_label_images(source_folder, dest_folder):  
    # 检查目标文件夹是否存在，不存在则创建  
    if not os.path.exists(dest_folder):  
        os.makedirs(dest_folder)  
    
    # 遍历所有子文件夹  
    for root, dirs, files in os.walk(source_folder):  
        for dir_name in dirs:  
            subfolder_path = os.path.join(root, dir_name)  
            label_image_path = os.path.join(subfolder_path, 'label.png')  
            if os.path.exists(label_image_path):  
                # 目标文件路径  
                dest_image_path = os.path.join(dest_folder, f"{dir_name}.png")  
                # 复制文件并重命名  
                shutil.copy2(label_image_path, dest_image_path)  
                print(f"Copied {label_image_path} to {dest_image_path}")  

if __name__ == "__main__":  
    source_folder = 'D:\labelme\output'  # 替换为json转出来的dataset文件路径
    dest_folder = 'D:\labelme\output_mask'  # 替换为目标文件夹路径
    copy_and_rename_label_images(source_folder, dest_folder)

# copy_and_rename_label_images('Json_dataset','mask_png')