Skip to content

Instantly share code, notes, and snippets.

@hsuRush
Last active June 1, 2020 13:49
Show Gist options
  • Save hsuRush/f6689a62a740f0cf3ffa3ff99a79b7df to your computer and use it in GitHub Desktop.
Save hsuRush/f6689a62a740f0cf3ffa3ff99a79b7df to your computer and use it in GitHub Desktop.
for darknet
import xml.etree.ElementTree as ET
import os
from os import listdir
from os.path import join
import random
path_to_dataset = '/data1000G/steven/lpDataSet/AOLP/all_train_sets/'
# images are at path_to_dataset + '/images/'
# the voc labels are at path_to_dataset + '/labels_voc/'
# the filename of images and labels should be exactly same.
# for example:
# path_to_dataset
# │
# └───images
# │ │
# │ │ file001.jpg
# │ │ file002.jpg
# │ └ ...
# │
# └───labels_voc
# │ │ file001.xml
# │ │ file002.xml
classes = ["plate"]
valid_split = 0.1
def convert(size, box):
dw = 1./size[0]
dh = 1./size[1]
x = (box[0] + box[1])/2.0
y = (box[2] + box[3])/2.0
w = box[1] - box[0]
h = box[3] - box[2]
x = x*dw
w = w*dw
y = y*dh
h = h*dh
return (x,y,w,h)
def convert_annotation(image_id):
in_file = open(path_to_dataset + 'labels_voc/%s.xml'%(image_id)) #輸入路徑
out_file = open(path_to_dataset + 'labels/%s.txt'%(image_id), 'w') #輸出路徑
tree=ET.parse(in_file) #得到xml樹
root = tree.getroot() #得到根
size = root.find('size') #通過size標籤得到尺寸信息
w = int(size.find('width').text) #分別得到照片的寬和高
h = int(size.find('height').text)
for obj in root.iter('object'): #查找到每一個標籤對象
difficult = obj.find('difficult').text #獲得difficult標籤內容
cls = obj.find('name').text #獲得name標籤內容
if cls not in classes or int(difficult) == 1:
continue
cls_id = classes.index(cls)
xmlbox = obj.find('bndbox') #獲取boundbox
b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text), float(xmlbox.find('ymin').text), float(xmlbox.find('ymax').text))
bb = convert((w,h), b)
out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n') #給bb中數之間添加空格,並輸出
if not os.path.exists(path_to_dataset + 'labels/'): #如果不存在要輸出的文件夾,則先創建
os.makedirs(path_to_dataset + 'labels/')
image_ids = [f[:-4] for f in os.listdir(path_to_dataset + 'images/') if os.path.isfile(os.path.join(path_to_dataset + 'images/', f)) and ('jpg' in f)]
for image_id in image_ids:
convert_annotation(image_id)
random.shuffle(image_ids) #打亂文件名
train_data_num = int((1-valid_split)*len(image_ids))
train_ids = image_ids[0:train_data_num] #取前384個為訓練集
train_list = open('train.txt', 'w') #打開要輸出的圖片路徑信息
for train_id in train_ids:
train_list.write(path_to_dataset + 'images/%s.jpg\n'%(train_id)) #將訓練圖片的路徑寫入文件中
train_list.close()
if not ((valid_split == 0.0) or (valid_split == 0)):
test_ids = image_ids[train_data_num:len(image_ids)] #剩餘的為測試集
test_list = open('test.txt', 'w') #打開要輸出的圖片路徑信息
for test_id in test_ids:
test_list.write(path_to_dataset + '/images/%s.jpg\n'%(test_id)) #將測試圖片的路徑寫入文件中
test_list.close()
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment