voc数据集的充分利用——将图片和xml按类别保存在不同文件夹、将目标剪裁后按类别保存在不同文件夹
前言：
在做深度学习的时候，经常需要收集样本，有些样本我们可以从开源数据库中提取，省去自己标注的麻烦，下面介绍几种提取的方法，根据自己需要拿去用。
1. 将图片按类别保存在不同文件夹，文件名不变。
执行完得到如下结果，只是对图片进行的分类，没有对xml进行分类。
对xml和图片都进行分类的代码参考本博客第3部分介绍。
voc_class-pic.py
import xml.dom.minidom
import os
import cv2
################
FindPath = './VOC2012/Annotations/'
FileNames = os.listdir(FindPath)
pic_path = './VOC2012/JPEGImages/'
save_path_pic = './VOC2012-class/'
Resnet_height = 224
Rsenet_width = 224
start_name = 0
one_location_list = []
all_location_list = []
all_name_list = []
def get_all_location(now_box_root):
for box_i in range(len(now_box_root)):
location_xmin = now_box[box_i].getElementsByTagName('xmin')
location_xmax = now_box[box_i].getElementsByTagName('xmax')
location_ymin = now_box[box_i].getElementsByTagName('ymin')
location_ymax = now_box[box_i].getElementsByTagName('ymax')
location_xmin = location_xmin[0].firstChild.data
location_xmax = location_xmax[0].firstChild.data
location_ymin = location_ymin[0].firstChild.data
location_ymax = location_ymax[0].firstChild.data
return location_xmin, location_xmax , location_ymin , location_ymax
def get_path(target_save_path):
target_path = save_path_pic + target_save_path + '/'
if os.path.exists(target_path) is False:
os.makedirs(target_path)
print('target_path = ',target_path)
return target_path
def crop_pic(start_name , picName , img_name ,location_size):
img = cv2.imread(pic_path + picName + '.jpg')
for img_i in range(len(img_name)):
print('1 = ',location_size[img_i][0] ,' ',location_size[img_i][1] ,' ' ,location_size[img_i][2], ' ',location_size[img_i][3])
image = img[ location_size[img_i][2]:location_size[img_i][3] , location_size[img_i][0]:location_size[img_i][1] ]
width = location_size[img_i][1] - location_size[img_i][0]
height = location_size[img_i][3] - location_size[img_i][2]
target_width = (Resnet_height * width) // height
#image = cv2.resize(image , (Resnet_height , Resnet_height) ,interpolation=cv2.INTER_CUBIC) #resize
crop_path = get_path(img_name[img_i])
print('crop_path = ',crop_path)
###### save crop pic
#cv2.imwrite(crop_path + picName + '.jpg',image)
###### save original pic
cv2.imwrite(crop_path + picName + '.jpg',img)
for file_name in FileNames:
dom = xml.dom.minidom.parse(os.path.join(FindPath, file_name))
# print('filename = ',file_name)
get_file_to_pic_name,err_xml = os.path.splitext(file_name)
print('---------------------------')
print('before = ',get_file_to_pic_name)
root = dom.documentElement
object_root = root.getElementsByTagName('object')
length = len(object_root)
for root_i in range(length):
now_name = object_root[root_i].getElementsByTagName('name')
now_box = object_root[root_i].getElementsByTagName('bndbox')
for get_name_nums in range(len(now_name)):
####### get name
get_object_name = now_name[get_name_nums].firstChild.data
print('get_name = ',get_object_name)
all_name_list.append(get_object_name)
####### get location
get_xmin , get_xmax , get_ymin , get_ymax = get_all_location(now_box)
one_location_list.append(int(get_xmin))
one_location_list.append(int(get_xmax))
one_location_list.append(int(get_ymin))
one_location_list.append(int(get_ymax))
all_location_list.append(one_location_list)
one_location_list = []
# print('all = ',all_location_list)
if len(all_name_list) != len(all_location_list):
print('Error file is ',file_name,',shut down!')
break
# print('len = ',len(all_name_list),' ',len(all_location_list))
############ crop pic
crop_pic(start_name , get_file_to_pic_name,all_name_list , all_location_list)
start_name += 1
all_name_list=[]
all_location_list=[]
2. 将图片剪裁后，按类别保存在不同的文件夹
执行完之后的结果如下，图片是剪裁后的：
代码和上面那段就一句不同，放在这里直接copy去用。
voc_crop.py
import xml.dom.minidom
import os
import cv2
################
FindPath = './VOC2012/Annotations/'
FileNames = os.listdir(FindPath)
pic_path = './VOC2012/JPEGImages/'
save_path_pic = './VOC2012-crop/'
Resnet_height = 224
Rsenet_width = 224
start_name = 0
one_location_list = []
all_location_list = []
all_name_list = []
def get_all_location(now_box_root):
for box_i in range(len(now_box_root)):
location_xmin = now_box[box_i].getElementsByTagName('xmin')
location_xmax = now_box[box_i].getElementsByTagName('xmax')
location_ymin = now_box[box_i].getElementsByTagName('ymin')
location_ymax = now_box[box_i].getElementsByTagName('ymax')
location_xmin = location_xmin[0].firstChild.data
location_xmax = location_xmax[0].firstChild.data
location_ymin = location_ymin[0].firstChild.data
location_ymax = location_ymax[0].firstChild.data
return location_xmin, location_xmax , location_ymin , location_ymax
def get_path(target_save_path):
target_path = save_path_pic + target_save_path + '/'
if os.path.exists(target_path) is False:
os.makedirs(target_path)
print('target_path = ',target_path)
return target_path
def crop_pic(start_name , picName , img_name ,location_size):
img = cv2.imread(pic_path + picName + '.jpg')
for img_i in range(len(img_name)):
print('1 = ',location_size[img_i][0] ,' ',location_size[img_i][1] ,' ' ,location_size[img_i][2], ' ',location_size[img_i][3])
image = img[ location_size[img_i][2]:location_size[img_i][3] , location_size[img_i][0]:location_size[img_i][1] ]
width = location_size[img_i][1] - location_size[img_i][0]
height = location_size[img_i][3] - location_size[img_i][2]
target_width = (Resnet_height * width) // height
#image = cv2.resize(image , (Resnet_height , Resnet_height) ,interpolation=cv2.INTER_CUBIC) #resize
crop_path = get_path(img_name[img_i])
print('crop_path = ',crop_path)
###### save crop pic
cv2.imwrite(crop_path + picName + '.jpg',image)
###### save original pic
#cv2.imwrite(crop_path + picName + '.jpg',img)
for file_name in FileNames:
dom = xml.dom.minidom.parse(os.path.join(FindPath, file_name))
# print('filename = ',file_name)
get_file_to_pic_name,err_xml = os.path.splitext(file_name)
print('---------------------------')
print('before = ',get_file_to_pic_name)
root = dom.documentElement
object_root = root.getElementsByTagName('object')
length = len(object_root)
for root_i in range(length):
now_name = object_root[root_i].getElementsByTagName('name')
now_box = object_root[root_i].getElementsByTagName('bndbox')
for get_name_nums in range(len(now_name)):
####### get name
get_object_name = now_name[get_name_nums].firstChild.data
print('get_name = ',get_object_name)
all_name_list.append(get_object_name)
####### get location
get_xmin , get_xmax , get_ymin , get_ymax = get_all_location(now_box)
one_location_list.append(int(get_xmin))
one_location_list.append(int(get_xmax))
one_location_list.append(int(get_ymin))
one_location_list.append(int(get_ymax))
all_location_list.append(one_location_list)
one_location_list = []
# print('all = ',all_location_list)
if len(all_name_list) != len(all_location_list):
print('Error file is ',file_name,',shut down!')
break
# print('len = ',len(all_name_list),' ',len(all_location_list))
############ crop pic
crop_pic(start_name , get_file_to_pic_name,all_name_list , all_location_list)
start_name += 1
all_name_list=[]
all_location_list=[]
3. 将voc数据集按类别保存图片，按类别保存xml标注文件。
执行之后，会将person相关的图片和xml都提取出来。
每次只能分出一种类别，例如“person”类别提取代码如下，要提取其他类别，需要修改代码，需要修改的地方我再下面注释了########### 1 change，########### 2 change，########### 3 change，另外路径根据自己的需要修改。
voc-class-pic-xml.py
import os
import os.path
import shutil
fileDir_ann = './VOC2012/Annotations/'
fileDir_img = './VOC2012/JPEGImages/'
########### 1 change
saveDir_img = './VOC2012-class-xml/person/'
if not os.path.exists(saveDir_img):
os.mkdir(saveDir_img)
names = locals()
for files in os.walk(fileDir_ann):
for file in files[2]:
print file + "-->start!"
########### 2 change
saveDir_ann = './VOC2012-class-xml/person/'
if not os.path.exists(saveDir_ann):
os.mkdir(saveDir_ann)
fp = open(fileDir_ann + file)
saveDir_ann = saveDir_ann + file
fp_w = open(saveDir_ann, 'w')
classes = ['aeroplane','bicycle','bird','boat','bottle','bus','car','cat','chair','cow','diningtable',\
'dog','horse','motorbike','pottedplant','sheep','sofa','train','tvmonitor','person']
lines = fp.readlines()
ind_start = []
ind_end = []
lines_id_start = lines[:]
lines_id_end = lines[:]
while "\t in lines_id_start:
a = lines_id_start.index("\t)
ind_start.append(a)
lines_id_start[a] = "delete"
while "\t\n" in lines_id_end:
b = lines_id_end.index("\t\n")
ind_end.append(b)
lines_id_end[b] = "delete"
for k in range(0,len(ind_start)):
for j in range(0,len(classes)):
if classes[j] in lines[ind_start[k]+1]:
a = ind_start[k]
names['block%d'%k] = lines[a:ind_end[k]+1]
break
########### 3 change
classes1 = '\t\tperson \n'
string_start = lines[0:ind_start[0]]
string_end = lines[ind_end[-1] + 1:]
a = 0
for k in range(0,len(ind_start)):
if classes1 in names['block%d'%k]:
a += 1
string_start += names['block%d'%k]
string_start += string_end
for c in range(0,len(string_start)):
fp_w.write(string_start[c])
fp_w.close()
if a == 0:
os.remove(saveDir_ann)
else:
name_img = fileDir_img + os.path.splitext(file)[0] + ".jpg"
shutil.copy(name_img,saveDir_img)
fp.close()
参考资料：
https://www.cnblogs.com/tyty-Somnuspoppy/p/10250486.html
https://download.csdn.net/download/u014513323/10823680