1、准备工作
目标检测数据的标注分为两种格式：
- xml 解释性标签，左上角+右下角的坐标
- txt 记事本文件，类别+x，y中心坐标+w，h的相对值
需要准备的数据有：
其中images为图像数据，labels为txt文本信息
注意：这里仅仅支持图像格式为 jpg 格式！！！
需要转换数据格式，可以参考这章：PYTHON 自动化办公：更改图片后缀_改变文件夹里面图片后缀名的pytorch代码-CSDN博客
classes.txt 文件需要自行准备，没有的话，只能遍历出来看看目标框的是什么：
关于目标检测任务中，YOLO(txt格式)标注文件的可视化_yolo标注可视化-CSDN博客
2、代码
如下：
import os
from PIL import Image
import shutil
from tqdm import tqdmdef main(IMG,LAB,CLS):images = [i.split('.jpg')[0] for i in os.listdir(IMG)] # 数据if os.path.exists('VOCdevkit'):shutil.rmtree('VOCdevkit')os.makedirs('VOCdevkit')os.makedirs('VOCdevkit/VOC2007')os.makedirs('VOCdevkit/VOC2007/JPEGImages')os.makedirs('VOCdevkit/VOC2007/Annotations')for img in tqdm(images):with open(os.path.join('VOCdevkit/VOC2007/Annotations',img+'.xml'), 'w') as xml_files:shutil.copy(os.path.join(IMG, img + '.jpg'),'VOCdevkit/VOC2007/JPEGImages')image = Image.open(os.path.join(IMG, img + '.jpg'))img_w, img_h = image.sizexml_files.write('<annotation>\n')xml_files.write(' <folder>folder</folder>\n')xml_files.write(f' <filename>{img}.jpg</filename>\n')xml_files.write(' <source>\n')xml_files.write(' <database>Unknown</database>\n')xml_files.write(' </source>\n')xml_files.write(' <size>\n')xml_files.write(f' <width>{img_w}</width>\n')xml_files.write(f' <height>{img_h}</height>\n')xml_files.write(f' <depth>3</depth>\n')xml_files.write(' </size>\n')xml_files.write(' <segmented>0</segmented>\n')with open(os.path.join(LAB,img + '.txt'), 'r') as f:# 以列表形式返回每一行lines = f.read().splitlines()for each_line in lines:line = each_line.split(' ')xml_files.write(' <object>\n')xml_files.write(f' <name>{CLS[int(line[0])]}</name>\n')xml_files.write(' <pose>Unspecified</pose>\n')xml_files.write(' <truncated>0</truncated>\n')xml_files.write(' <difficult>0</difficult>\n')xml_files.write(' <bndbox>\n')center_x = round(float(line[1]) * img_w)center_y = round(float(line[2]) * img_h)bbox_w = round(float(line[3]) * img_w)bbox_h = round(float(line[4]) * img_h)xmin = str(int(center_x - bbox_w / 2))ymin = str(int(center_y - bbox_h / 2))xmax = str(int(center_x + bbox_w / 2))ymax = str(int(center_y + bbox_h / 2))xml_files.write(f' <xmin>{xmin}</xmin>\n')xml_files.write(f' <ymin>{ymin}</ymin>\n')xml_files.write(f' <xmax>{xmax}</xmax>\n')xml_files.write(f' <ymax>{ymax}</ymax>\n')xml_files.write(' </bndbox>\n')xml_files.write(' </object>\n')xml_files.write('</annotation>')if __name__ == '__main__':# images 需要是 jpg格式yoloImages = 'yolodata/images'yoloLabels = 'yolodata/labels'classTxt = 'classes.txt'txt = open(classTxt, 'r')txt = txt.read().splitlines()print(txt)main(IMG=yoloImages,LAB=yoloLabels,CLS=txt)
效果如下：
3、其他代码
可视化代码可以参考：
from lxml import etree
import cv2# 读取 xml 文件信息，并返回字典形式
def parse_xml_to_dict(xml):if len(xml) == 0: # 遍历到底层，直接返回 tag对应的信息return {xml.tag: xml.text}result = {}for child in xml:child_result = parse_xml_to_dict(child) # 递归遍历标签信息if child.tag != 'object':result[child.tag] = child_result[child.tag]else:if child.tag not in result: # 因为object可能有多个，所以需要放入列表里result[child.tag] = []result[child.tag].append(child_result[child.tag])return {xml.tag: result}# xml 标注文件的可视化
def xmlShow(img, xml, save=True):image = cv2.imread(img)with open(xml, encoding='gb18030', errors='ignore') as fid: # 防止出现非法字符报错xml_str = fid.read()xml = etree.fromstring(xml_str)data = parse_xml_to_dict(xml)["annotation"] # 读取 xml文件信息ob = [] # 存放目标信息for i in data['object']: # 提取检测框name = str(i['name']) # 检测的目标类别bbox = i['bndbox']xmin = int(bbox['xmin'])ymin = int(bbox['ymin'])xmax = int(bbox['xmax'])ymax = int(bbox['ymax'])tmp = [name, xmin, ymin, xmax, ymax] # 单个检测框ob.append(tmp)# 绘制检测框for name, x1, y1, x2, y2 in ob:cv2.rectangle(image, (x1, y1), (x2, y2), color=(255, 0, 0), thickness=2) # 绘制矩形框cv2.putText(image, name, (x1, y1 - 10), fontFace=cv2.FONT_HERSHEY_SIMPLEX,fontScale=0.5, thickness=1, color=(0, 0, 255))# 保存图像if save:cv2.imwrite('result.png', image)# 展示图像cv2.imshow('test', image)cv2.waitKey()cv2.destroyAllWindows()if __name__ == "__main__":import osimport randomimg_path = r'.\VOCdevkit\VOC2007\JPEGImages' # 传入图片image = [os.path.join(img_path, i) for i in os.listdir(img_path)]r = random.randint(0, len(image) - 1) # 生成随机索引image_path = image[r]labels_path = image_path.replace('JPEGImages', 'Annotations') # 自动获取对应的 xml 标注文件labels_path = labels_path.replace('.jpg', '.xml')xmlShow(img=image_path, xml=labels_path, save=True)
至于xml生成标签的json字典文件，参考：
目标检测篇：如何根据xml标注文件生成类别classes的json文件_将xml 根据注解 解读 成类-CSDN博客