1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
|
import os
import json
import threading
import numpy as np
from PIL import Image
import tensorflow as tf
from keras import losses
from keras import backend as K
from keras.utils import plot_model
from keras.preprocessing import image
from keras.preprocessing.sequence import pad_sequences
from keras.layers import Input, Dense, Flatten
from keras.layers.core import Reshape, Masking, Lambda, Permute
from keras.layers.recurrent import GRU, LSTM
from keras.layers.wrappers import Bidirectional, TimeDistributed
from keras.layers.normalization import BatchNormalization
from keras.layers.convolutional import Conv2D, MaxPooling2D, ZeroPadding2D
from keras.optimizers import SGD, Adam
from keras.models import Model
from keras.callbacks import EarlyStopping, ModelCheckpoint, LearningRateScheduler, TensorBoard
from imp import reload
import model
import io
import sys
reload(sys)
img_h = 32
img_w = 280
batch_size = 128
maxlabellength = 10
def get_session(gpu_fraction=1.0):
num_threads = os.environ.get('OMP_NUM_THREADS')
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=gpu_fraction)
if num_threads:
return tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options, intra_op_parallelism_threads=num_threads))
else:
return tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
def readfile(filename):
res = []
with open(filename, 'r') as f:
lines = f.readlines()
for i in lines:
res.append(i.strip())
dic = {}
for i in res:
p = i.split(' ')
dic[p[0]] = p[1:]
return dic
class random_uniform_num():
"""
均匀随机，确保每轮每个只出现一次
"""
def __init__(self, total):
self.total = total
self.range = [i for i in range(total)]
np.random.shuffle(self.range)
self.index = 0
def get(self, batchsize):
r_n=[]
if(self.index + batchsize > self.total):
r_n_1 = self.range[self.index:self.total]
np.random.shuffle(self.range)
self.index = (self.index + batchsize) - self.total
r_n_2 = self.range[0:self.index]
r_n.extend(r_n_1)
r_n.extend(r_n_2)
else:
r_n = self.range[self.index : self.index + batchsize]
self.index = self.index + batchsize
return r_n
def gen(data_file, image_path, batchsize=128, maxlabellength=10, imagesize=(32, 280)):
image_label = readfile(data_file)
_imagefile = [i for i, j in image_label.items()]
x = np.zeros((batchsize, imagesize[0], imagesize[1], 1), dtype=np.float)
labels = np.ones([batchsize, maxlabellength]) * 10000
input_length = np.zeros([batchsize, 1])
label_length = np.zeros([batchsize, 1])
r_n = random_uniform_num(len(_imagefile))
_imagefile = np.array(_imagefile)
while 1:
shufimagefile = _imagefile[r_n.get(batchsize)]
for i, j in enumerate(shufimagefile):
img1 = Image.open(os.path.join(image_path, j)).convert('L')
img = np.array(img1, 'f') / 255.0 - 0.5
x[i] = np.expand_dims(img, axis=2)
str = image_label[j]
label_length[i] = len(str)
if(len(str) <= 0):
print("len < 0", j)
input_length[i] = imagesize[1] // 8
labels[i, :len(str)] = [int(k) - 1 for k in str]
inputs = {'the_input': x,
'the_labels': labels,
'input_length': input_length,
'label_length': label_length,
}
outputs = {'ctc': np.zeros([batchsize])}
yield (inputs, outputs)
def ctc_lambda_func(args):
y_pred, labels, input_length, label_length = args
return K.ctc_batch_cost(labels, y_pred, input_length, label_length)
def get_model(img_h, nclass):
input = Input(shape=(img_h, None, 1), name='the_input')
y_pred = model.dense_cnn(input, nclass)
basemodel = Model(inputs=input, outputs=y_pred)
basemodel.summary()
labels = Input(name='the_labels', shape=[None], dtype='float32')
input_length = Input(name='input_length', shape=[1], dtype='int64')
label_length = Input(name='label_length', shape=[1], dtype='int64')
loss_out = Lambda(ctc_lambda_func, output_shape=(1,), name='ctc')([y_pred, labels, input_length, label_length])
model = Model(inputs=[input, labels, input_length, label_length], outputs=loss_out)
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer='adam', metrics=['accuracy'])
return basemodel, model
if __name__ == '__main__':
char_set = io.open('char_std_5990.txt', 'r', encoding='utf-8').readlines()
char_set = ''.join([ch.strip('\n') for ch in char_set][1:])
nclass = len(char_set)
K.set_session(get_session())
reload(model)
basemodel, model = get_model(img_h, nclass)
modelPath = './models/pretrain_model/keras.h5'
if os.path.exists(modelPath):
print("Loading model weights...")
basemodel.load_weights(modelPath)
print('done!')
train_loader = gen('data_train.txt', './images', batchsize=batch_size, maxlabellength=maxlabellength, imagesize=(img_h, img_w))
test_loader = gen('data_test.txt', './images', batchsize=batch_size, maxlabellength=maxlabellength, imagesize=(img_h, img_w))
checkpoint = ModelCheckpoint(filepath='./models/model-weights-{epoch:02d}-{val_loss:.2f}.h5', monitor='val_loss', save_best_only=False, save_weights_only=True)
lr_schedule = lambda epoch: 0.0005 * 0.4**epoch
learning_rate = np.array([lr_schedule(i) for i in range(10)])
changelr = LearningRateScheduler(lambda epoch: float(learning_rate[epoch]))
earlystop = EarlyStopping(monitor='val_loss', patience=2, verbose=1)
tensorboard = TensorBoard(log_dir='./models/logs', write_graph=True)
print('-----------Start training-----------')
model.fit_generator(train_loader,
steps_per_epoch = 3607567 // batch_size,
epochs = 10,
initial_epoch = 0,
validation_data = test_loader,
validation_steps = 36440 // batch_size,
callbacks = [checkpoint, earlystop, changelr, tensorboard])
|
