目录
lxml windows安装
读取示例
可视化
生成示例
上面是代码&＃xff0c;下面有调用示例
api调用代码&＃xff0c;其实只有几行&＃xff1a;
这个生成代码也很简洁&＃xff1a;
python生成voc xml文件_jacke121的专栏-CSDN博客
lxml windows安装
pip install lxml可能失败&＃xff0c;
一种可行方法&＃xff0c;在这里下载对应版本whl&＃xff0c;然后安装&＃xff0c;亲测OK。
lxml · PyPI
读取示例
import os
import cv2
import numpy as np
import xml.etree.ElementTree as ET
CLASSES &＃61; (&＃39;m&＃39;,)
_class_to_ind&＃61;dict(list(zip(CLASSES, list(range(len(CLASSES))))))
def get_label(label_path,shape, flip&＃61;False):h, w, c &＃61; shapeannotations &＃61; np.zeros((0, 5),dtype&＃61;np.float32)if os.path.exists(label_path):tree &＃61; ET.parse(label_path)objs &＃61; tree.findall(&＃39;object&＃39;)for ix, obj in enumerate(objs):if obj.find(&＃39;name&＃39;).text.lower().strip() not in _class_to_ind:# logging.info("label not select: {}".format(obj.find(&＃39;name&＃39;).text.lower().strip()))continuebbox &＃61; obj.find(&＃39;bndbox&＃39;)x1 &＃61; max(float(bbox.find(&＃39;xmin&＃39;).text), 1) # - 1y1 &＃61; max(float(bbox.find(&＃39;ymin&＃39;).text), 1) # - 1x2 &＃61; min(float(bbox.find(&＃39;xmax&＃39;).text), 10000) # - 1y2 &＃61; min(float(bbox.find(&＃39;ymax&＃39;).text), 10000) # - 1if obj.find("difficult") is not None and obj.find("difficult").text &＃61;&＃61; &＃39;1&＃39;:if (x2-x1) * (y2-y1)<10*10:continue
可视化
import os
import cv2
import reimport xml.etree.ElementTree as ETCLASSES &＃61; (&＃39;car&＃39;,)
_class_to_ind&＃61;dict(list(zip(CLASSES, list(range(len(CLASSES))))))if __name__ &＃61;&＃61; &＃39;__main__&＃39;:image_dir &＃61; r&＃39;G:\img&＃39;xml_dir &＃61; r&＃39;G:\xml&＃39;image_list &＃61; os.listdir(image_dir)cnt &＃61; 0for i in image_list:image_path &＃61; os.path.join(image_dir ,i)xml_path &＃61; os.path.join(xml_dir ,os.path.splitext(i)[0 ] &＃43;&＃39;.xml&＃39;)tree &＃61; ET.parse(xml_path)objs &＃61; tree.findall(&＃39;object&＃39;)image &＃61; cv2.imread(image_path)for ix, obj in enumerate(objs):if obj.find(&＃39;name&＃39;).text.lower().strip() not in _class_to_ind:# logging.info("label not select: {}".format(obj.find(&＃39;name&＃39;).text.lower().strip()))continuebbox &＃61; obj.find(&＃39;bndbox&＃39;)x1 &＃61; max(float(bbox.find(&＃39;xmin&＃39;).text), 1) # - 1y1 &＃61; max(float(bbox.find(&＃39;ymin&＃39;).text), 1) # - 1x2 &＃61; min(float(bbox.find(&＃39;xmax&＃39;).text), 10000) # - 1y2 &＃61; min(float(bbox.find(&＃39;ymax&＃39;).text), 10000) # - 1cv2.rectangle(image, (int(x1), int(y1)), (int(x2), int(y2)), (255, 0, 0), thickness&＃61;2)# cv2.putText(image, info[0], (info[1], info[2]), cv2.FONT_HERSHEY_PLAIN, 0.5, (0, 0, 255), 1)cv2.imshow("image",image)cv2.waitKey()
生成示例
上面是代码&＃xff0c;下面有调用示例
api调用代码&＃xff0c;其实只有几行&＃xff1a;
anno &＃61; GEN_Annotations(pic_path)anno.set_size(new_width, new_height, 3)for (x, y, w, h,name) in bbox_list:anno.add_pic_attr(name, x, y, w, h)save_xml_path&＃61;xml_path.replace(base_path,save_pic_path)os.makedirs(os.path.dirname(save_xml_path), exist_ok&＃61;True)anno.savefile(save_xml_path)
# -*- coding: utf-8 -*-import osfrom lxml import etreeclass GEN_Annotations:def __init__(self, filename):self.root &＃61; etree.Element("annotation")child1 &＃61; etree.SubElement(self.root, "folder")child1.text &＃61; "VOC2007"child2 &＃61; etree.SubElement(self.root, "filename")child2.text &＃61; filenamechild3 &＃61; etree.SubElement(self.root, "source")# child2.set("database", "The VOC2007 Database")child4 &＃61; etree.SubElement(child3, "annotation")child4.text &＃61; "PASCAL VOC2007"child5 &＃61; etree.SubElement(child3, "database")child6 &＃61; etree.SubElement(child3, "image")child6.text &＃61; "flickr"child7 &＃61; etree.SubElement(child3, "flickrid")child7.text &＃61; "35435"# root.append( etree.Element("child1") )# root.append( etree.Element("child1", interesting&＃61;"totally"))# child2 &＃61; etree.SubElement(root, "child2")# child3 &＃61; etree.SubElement(root, "child3")# root.insert(0, etree.Element("child0"))def set_size(self,witdh,height,channel):size &＃61; etree.SubElement(self.root, "size")widthn &＃61; etree.SubElement(size, "width")widthn.text &＃61; str(witdh)heightn &＃61; etree.SubElement(size, "height")heightn.text &＃61; str(height)channeln &＃61; etree.SubElement(size, "channel")channeln.text &＃61; str(channel)def savefile(self,filename):tree &＃61; etree.ElementTree(self.root)tree.write(filename, pretty_print&＃61;True, xml_declaration&＃61;False, encoding&＃61;&＃39;utf-8&＃39;)def add_pic_attr(self,label,x,y,w,h,diffi_&＃61;0):object &＃61; etree.SubElement(self.root, "object")namen &＃61; etree.SubElement(object, "name")namen.text &＃61; labeldiffi &＃61; etree.SubElement(object, "difficult")diffi.text &＃61; str(diffi_)truncated_ &＃61; etree.SubElement(object, "truncated")truncated_.text &＃61; str(0)bndbox &＃61; etree.SubElement(object, "bndbox")xminn &＃61; etree.SubElement(bndbox, "xmin")xminn.text &＃61; str(x)yminn &＃61; etree.SubElement(bndbox, "ymin")yminn.text &＃61; str(y)xmaxn &＃61; etree.SubElement(bndbox, "xmax")xmaxn.text &＃61; str(x&＃43;w)ymaxn &＃61; etree.SubElement(bndbox, "ymax")ymaxn.text &＃61; str(y&＃43;h)
调用&＃xff1a;
import cv2
import xml.etree.ElementTree as ETdef change_size(pic_path, xml_path,base_path,save_pic_path):img &＃61; cv2.imread(pic_path)if img is None:print(&＃39;img is null&＃39;, pic_path)returnif img.shape[1] <1 or img.shape[0] <1:print(&＃39;img is null&＃39;, pic_path)returnheight,width&＃61;img.shape[:2]new_width &＃61; widthnew_height &＃61; heightif height>700 and width>1200:new_width &＃61; int(width * 0.8)new_height &＃61; int(height * 0.8)img&＃61;cv2.resize(img,(new_width,new_height))if new_height!&＃61;576 or new_width!&＃61;1024:print(new_height,new_width,pic_path)pic_path &＃61; pic_path.replace(base_path, save_pic_path)os.makedirs(os.path.dirname(pic_path), exist_ok&＃61;True)cv2.imwrite(pic_path, img)if os.path.exists(xml_path):tree &＃61; ET.parse(xml_path)# file_objs &＃61; tree.findall(&＃39;filename&＃39;)objs &＃61; tree.findall(&＃39;object&＃39;)objs_diffi &＃61; tree.findall(&＃39;object/difficult&＃39;)if len(objs_diffi)&＃61;&＃61;0:print(xml_path, objs_diffi, len(objs_diffi))bbox_list &＃61; []for ix, obj in enumerate(objs):name &＃61; obj.find(&＃39;name&＃39;).textbbox &＃61; obj.find(&＃39;bndbox&＃39;)x1 &＃61; int(int(bbox.find(&＃39;xmin&＃39;).text)*0.8)y1 &＃61; int(int(bbox.find(&＃39;ymin&＃39;).text)*0.8)x2 &＃61; int(int(bbox.find(&＃39;xmax&＃39;).text)*0.8)y2 &＃61; int(int(bbox.find(&＃39;ymax&＃39;).text)*0.8)bbox_list.append((x1,y1,(x2-x1),(y2-y1),name))anno &＃61; GEN_Annotations(pic_path)anno.set_size(new_width, new_height, 3)for (x, y, w, h,name) in bbox_list:anno.add_pic_attr(name, x, y, w, h)save_xml_path&＃61;xml_path.replace(base_path,save_pic_path)os.makedirs(os.path.dirname(save_xml_path), exist_ok&＃61;True)anno.savefile(save_xml_path)if __name__ &＃61;&＃61; &＃39;__main__&＃39;:# save file pathsource_base_path &＃61; r&＃39;D:\Team-CV\dataset\chumao_train/&＃39;save_base_path &＃61; r&＃39;D:\Team-CV\dataset\chumao_train_1024/&＃39;os.makedirs(save_base_path, exist_ok&＃61;True)# file pathg &＃61; os.walk(source_base_path)pic_files &＃61; [&＃39;%s\\%s&＃39; % (i[0], j) for i in g if i[0].endswith(&＃39;JPEGImages&＃39;) for j in i[-1] if j.endswith(&＃39;jpg&＃39;)]for i,pic_file in enumerate(pic_files):if &＃39;JPEGImages&＃39; in pic_file:xml_file &＃61; pic_file.replace(&＃39;JPEGImages&＃39;,&＃39;Annotations&＃39;).replace(&＃39;.jpg&＃39;,&＃39;.xml&＃39;)change_size(pic_file, xml_file,source_base_path,save_base_path)print(i,pic_file)
京公网安备 11010802041100号