1 #  write in tfrecord
  2 import tensorflow as tf
  3 import os
  4 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
  5
  6
  7 FLAGS = tf.app.flags.FLAGS
  8 tf.app.flags.DEFINE_string("tfrecords_dir", "./tfrecords/captcha.tfrecords", "验证码tfrecords文件")
  9 tf.app.flags.DEFINE_string("captcha_dir", "../data/Genpics/", "验证码图片路径")
 10 tf.app.flags.DEFINE_string("letter", "ABCDEFGHIJKLMNOPQRSTUVWXYZ", "验证码字符的种类")
 11
 12
 13 def dealwithlabel(label_str):
 14
 15     # 构建字符索引 {0：'A', 1:'B'......}
 16     num_letter = dict(enumerate(list(FLAGS.letter)))
 17
 18     # 键值对反转 {'A':0, 'B':1......}
 19     letter_num = dict(zip(num_letter.values(), num_letter.keys()))
 20
 21     print(letter_num)
 22
 23     # 构建标签的列表
 24     array = []
 25
 26     # 给标签数据进行处理[[b"NZPP"]......]
 27     for string in label_str:
 28
 29         letter_list = []# [1,2,3,4]
 30
 31         # 修改编码，bytes --> string
 32         for letter in string.decode('utf-8'):
 33             letter_list.append(letter_num[letter])
 34
 35         array.append(letter_list)
 36
 37     # [[13, 25, 15, 15], [22, 10, 7, 10], [22, 15, 18, 9], [16, 6, 13, 10], [1, 0, 8, 17], [0, 9, 24, 14].....]
 38     print(array)
 39
 40     # 将array转换成tensor类型
 41     label = tf.constant(array)
 42
 43     return label
 44
 45
 46 def get_captcha_image():
 47     """
 48     获取验证码图片数据
 49     :param file_list: 路径+文件名列表
 50     :return: image
 51     """
 52     # 构造文件名
 53     filename = []
 54
 55     for i in range(6000):
 56         string = str(i) + ".jpg"
 57         filename.append(string)
 58
 59     # 构造路径+文件
 60     file_list = [os.path.join(FLAGS.captcha_dir, file) for file in filename]
 61
 62     # 构造文件队列
 63     file_queue = tf.train.string_input_producer(file_list, shuffle=False)
 64
 65     # 构造阅读器
 66     reader = tf.WholeFileReader()
 67
 68     # 读取图片数据内容
 69     key, value = reader.read(file_queue)
 70
 71     # 解码图片数据
 72     image = tf.image.decode_jpeg(value)
 73
 74     image.set_shape([20, 80, 3])
 75
 76     # 批处理数据 [6000, 20, 80, 3]
 77     image_batch = tf.train.batch([image], batch_size=6000, num_threads=1, capacity=6000)
 78
 79     return image_batch
 80
 81
 82 def get_captcha_label():
 83     """
 84     读取验证码图片标签数据
 85     :return: label
 86     """
 87     file_queue = tf.train.string_input_producer(["../data/Genpics/labels.csv"], shuffle=False)
 88
 89     reader = tf.TextLineReader()
 90
 91     key, value = reader.read(file_queue)
 92
 93     records = [[1], ["None"]]
 94
 95     number, label = tf.decode_csv(value, record_defaults=records)
 96
 97     # [["NZPP"], ["WKHK"], ["ASDY"]]
 98     label_batch = tf.train.batch([label], batch_size=6000, num_threads=1, capacity=6000)
 99
100     return label_batch
101
102
103 def write_to_tfrecords(image_batch, label_batch):
104     """
105     将图片内容和标签写入到tfrecords文件当中
106     :param image_batch: 特征值
107     :param label_batch: 标签纸
108     :return: None
109     """
110     # 转换类型
111     label_batch = tf.cast(label_batch, tf.uint8)
112
113     print(label_batch)
114
115     # 建立TFRecords 存储器
116     writer = tf.python_io.TFRecordWriter(FLAGS.tfrecords_dir)
117
118     # 循环将每一个图片上的数据构造example协议块，序列化后写入
119     for i in range(6000):
120         # 取出第i个图片数据，转换相应类型,图片的特征值要转换成字符串形式
121         image_string = image_batch[i].eval().tostring()
122
123         # 标签值，转换成整型
124         label_string = label_batch[i].eval().tostring()
125
126         # 构造协议块
127         example = tf.train.Example(features=tf.train.Features(feature={
128             "image": tf.train.Feature(bytes_list=tf.train.BytesList(value=[image_string])),
129             "label": tf.train.Feature(bytes_list=tf.train.BytesList(value=[label_string]))
130         }))
131
132         writer.write(example.SerializeToString())
133
134     # 关闭文件
135     writer.close()
136
137     return None
138
139
140 if __name__ == "__main__":
141
142     # 获取验证码文件当中的图片
143     image_batch = get_captcha_image()
144
145     # 获取验证码文件当中的标签数据
146     label = get_captcha_label()
147
148     print(image_batch, label)
149
150     with tf.Session() as sess:
151
152         coord = tf.train.Coordinator()
153
154         threads = tf.train.start_queue_runners(sess=sess, coord=coord)
155
156         # 获取tensor里面的值
157         label_str = sess.run(label)
158
159         print(label_str)
160
161         # 处理字符串标签到数字张量
162         label_batch = dealwithlabel(label_str)
163
164         print(label_batch)
165
166         # 将图片数据和内容写入到tfrecords文件当中
167         write_to_tfrecords(image_batch, label_batch)
168
169         coord.request_stop()
170
171         coord.join(threads)

 1 # read tfrecords
 2 def read_and_decode():
 3     """
 4     读取验证码数据API
 5     :return: image_batch, label_batch
 6     """
 7     # 1、构建文件队列
 8     file_queue = tf.train.string_input_producer([FLAGS.captcha_dir])
 9
10     # 2、构建阅读器，读取文件内容，默认一个样本
11     reader = tf.TFRecordReader()
12
13     # 读取内容
14     key, value = reader.read(file_queue)
15
16     # tfrecords格式example,需要解析
17     features = tf.parse_single_example(value, features={
18         "image": tf.FixedLenFeature([], tf.string),
19         "label": tf.FixedLenFeature([], tf.string),
20     })
21
22     # 解码内容，字符串内容
23     # 1、先解析图片的特征值
24     image = tf.decode_raw(features["image"], tf.uint8)
25     # 1、先解析图片的目标值
26     label = tf.decode_raw(features["label"], tf.uint8)
27
28     # print(image, label)
29
30     # 改变形状
31     image_reshape = tf.reshape(image, [20, 80, 3])
32
33     label_reshape = tf.reshape(label, [4])
34
35     print(image_reshape, label_reshape)
36
37     # 进行批处理,每批次读取的样本数 100, 也就是每次训练时候的样本
38     image_batch, label_btach = tf.train.batch([image_reshape, label_reshape], batch_size=FLAGS.batch_size, num_threads=1, capacity=FLAGS.batch_size)
39
40     print(image_batch, label_btach)
41     return image_batch, label_btach

# write flags
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_string("tfrecords_dir", "./tfrecords/captcha.tfrecords", "验证码tfrecords文件")
tf.app.flags.DEFINE_string("captcha_dir", "../data/Genpics/", "验证码图片路径")
tf.app.flags.DEFINE_string("letter", "ABCDEFGHIJKLMNOPQRSTUVWXYZ", "验证码字符的种类")
# read flags
tf.app.flags.DEFINE_string("captcha_dir", "./tfrecords/captcha.tfrecords", "验证码数据的路径")
tf.app.flags.DEFINE_integer("batch_size", 100, "每批次训练的样本数")
tf.app.flags.DEFINE_integer("label_num", 4, "每个样本的目标值数量")
tf.app.flags.DEFINE_integer("letter_num", 26, "每个目标值取的字母的可能心个数")