-
Notifications
You must be signed in to change notification settings - Fork 30
/
model_recog_wrap.py
453 lines (427 loc) · 19.2 KB
/
model_recog_wrap.py
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
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
# -*- coding: utf-8 -*-
"""
@author: limingfan
"""
import os
import time
import random
from PIL import Image
import numpy as np
import tensorflow as tf
from tensorflow.python.framework import graph_util
import model_recog_def as model_def
import model_recog_meta as meta
#
TRAINING_STEPS = 16000
BATCH_SIZE = 128
#
LEARNING_RATE_BASE = 0.001
MOMENTUM = 0.9
REG_LAMBDA = 0.0001
GRAD_CLIP = 5.0
#
BATCH_SIZE_VALID = 1
VALID_FREQ = 100
LOSS_FREQ = 1
#
class ModelRecog():
#
HEIGHT_NORM = meta.height_norm
#
def __init__(self):
#
# default pb path
self.pb_file = os.path.join(meta.model_recog_dir, meta.model_recog_pb_file)
#
self.sess_config = tf.ConfigProto()
# self.sess_config.gpu_options.per_process_gpu_memory_fraction = 0.95
#
self.is_train = False
#
self.graph = None
self.sess = None
#
self.train_steps = TRAINING_STEPS
self.batch_size = BATCH_SIZE
#
self.learning_rate_base = LEARNING_RATE_BASE
self.momentum = MOMENTUM
self.reg_lambda = REG_LAMBDA
self.grad_clip = GRAD_CLIP
#
self.batch_size_valid = BATCH_SIZE_VALID
self.valid_freq = VALID_FREQ
self.loss_freq = LOSS_FREQ
#
def prepare_for_prediction(self, pb_file_path = None):
#
if pb_file_path == None: pb_file_path = self.pb_file
#
if not os.path.exists(pb_file_path):
print('ERROR: %s NOT exists, when load_pb_for_predict()' % pb_file_path)
return -1
#
self.graph = tf.Graph()
#
with self.graph.as_default():
#
with open(pb_file_path, "rb") as f:
graph_def = tf.GraphDef()
graph_def.ParseFromString(f.read())
#
tf.import_graph_def(graph_def, name="")
#
#
# input/output variables
#
self.x = self.graph.get_tensor_by_name('x-input:0')
self.w = self.graph.get_tensor_by_name('w-input:0')
#
self.seq_len = self.graph.get_tensor_by_name('seq_len:0')
self.result_logits = self.graph.get_tensor_by_name('rnn_logits/BiasAdd:0')
#
self.result_i = self.graph.get_tensor_by_name('CTCBeamSearchDecoder:0')
self.result_v = self.graph.get_tensor_by_name('CTCBeamSearchDecoder:1')
self.result_s = self.graph.get_tensor_by_name('CTCBeamSearchDecoder:2')
#
#
print('graph loaded for prediction')
#
self.sess = tf.Session(graph = self.graph, config = self.sess_config)
#
def predict(self, image_in):
#
# input data
if isinstance(image_in, str):
img = Image.open(image_in)
img = img.convert('RGB')
img_size = img.size
if img_size[1] != meta.height_norm:
w = int(img_size[0] * meta.height_norm *1.0/img_size[1])
img = img.resize((w, meta.height_norm))
img_data = np.array(img, dtype = np.float32)/255 # (height, width, channel)
img_data = [ img_data[:,:,0:3] ]
else:
# np array
img_data = image_in
#
w_arr = [ img_data[0].shape[1] ] # batch, height, width, channel
#
with self.graph.as_default():
#
feed_dict = {self.x: img_data, self.w: w_arr}
#
results, seq_length, d_i, d_v, d_s = \
self.sess.run([self.result_logits, self.seq_len,
self.result_i, self.result_v, self.result_s], feed_dict)
#
decoded = tf.SparseTensorValue(indices = d_i, values = d_v, dense_shape = d_s)
trans = model_def.convert2ListLabels(decoded)
#print(trans)
#
for item in trans:
seq = list(map(meta.mapOrder2Char, item))
str_result = ''.join(seq)
#
#
return str_result
#
def create_graph_all(self, training):
#
self.is_train = training
self.graph = tf.Graph()
#
with self.graph.as_default():
#
self.x = tf.placeholder(tf.float32, (None, None, None, 3), name = 'x-input')
self.w = tf.placeholder(tf.int32, (None,), name = 'w-input') # width
#
self.conv_feat, self.seq_len = model_def.conv_feat_layers(self.x, self.w, self.is_train) # train
self.result_logits = model_def.rnn_recog_layers(self.conv_feat, self.seq_len, len(meta.alphabet) + 1)
self.result_decoded_list = model_def.decode_rnn_results_ctc_beam(self.result_logits, self.seq_len)
#
print()
print('self.result_logits.op.name: ' + self.result_logits.op.name) # recog_logits/Sigmoid /BiasAdd
#print(self.seq_len) # seq_len
print('self.result_decoded_list[0]: ')
print(self.result_decoded_list[0]) # CTCBeamSearchDecoder
print()
#
# SparseTensor(indices=Tensor("CTCBeamSearchDecoder:0", shape=(?, 2), dtype=int64),
# values=Tensor("CTCBeamSearchDecoder:1", shape=(?,), dtype=int64),
# dense_shape=Tensor("CTCBeamSearchDecoder:2", shape=(2,), dtype=int64))
#
self.y = tf.sparse_placeholder(tf.int32, name = 'y-input')
#
print('self.y: ')
print(self.y)
#
# <tf.Operation 'y-input/shape' type=Placeholder>,
# <tf.Operation 'y-input/values' type=Placeholder>,
# <tf.Operation 'y-input/indices' type=Placeholder>]
#
#print(graph.get_operations())
#
self.loss = model_def.ctc_loss_layer(self.y, self.result_logits, self.seq_len)
#
#print(self.loss.op.name) # loss
#
# train
self.global_step = tf.train.get_or_create_global_step()
self.learning_rate = tf.get_variable("learning_rate", shape=[], dtype=tf.float32, trainable=False)
#optimizer = tf.train.MomentumOptimizer(learning_rate, MOMENTUM, use_nesterov=True)
#optimizer = tf.train.AdadeltaOptimizer(learning_rate=self.lr, epsilon=1e-6)
optimizer = tf.train.AdamOptimizer(learning_rate = self.learning_rate, beta1 = MOMENTUM)
#
'''
l2_loss = tf.add_n([tf.nn.l2_loss(var) for var in tf.trainable_variables()])
grads = optimizer.compute_gradients(self.loss + l2_loss * REG_LAMBDA)
gradients, variables = zip(*grads)
capped_grads, _ = tf.clip_by_global_norm(gradients, GRAD_CLIP)
grads_applying = zip(capped_grads, variables)
'''
grads_applying = optimizer.compute_gradients(self.loss)
self.train_op = optimizer.apply_gradients(grads_applying, global_step=self.global_step)
#
'''
#
# Update batch norm stats [http://stackoverflow.com/questions/43234667]
extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
#
with tf.control_dependencies(extra_update_ops):
#
learning_rate = tf.train.exponential_decay(LEARNING_RATE_BASE,
tf.train.get_global_step(),
DECAY_STEPS,
DECAY_RATE,
staircase = DECAY_STAIRCASE,
name = 'learning_rate')
optimizer = tf.train.AdamOptimizer(learning_rate = learning_rate,
beta1 = MOMENTUM)
# train_op =
train_op = tf.contrib.layers.optimize_loss(loss = loss,
global_step = tf.train.get_global_step(),
learning_rate = learning_rate,
optimizer = optimizer,
name = 'train_op') #variables = rnn_vars)
#
'''
#
print('global_step.op.name: ' + self.global_step.op.name)
print('train_op.name: ' + self.train_op.name)
print()
#
print('graph defined for training') if self.is_train else print('graph defined for validation')
#
#
def train_and_valid(self, data_train, data_valid):
#
# model save-path
if not os.path.exists(meta.model_recog_dir): os.mkdir(meta.model_recog_dir)
#
# graph
self.create_graph_all(training = True)
#
# restore and train
with self.graph.as_default():
#
saver = tf.train.Saver()
with tf.Session(config = self.sess_config) as sess:
#
tf.global_variables_initializer().run()
sess.run(tf.assign(self.learning_rate, tf.constant(self.learning_rate_base, dtype=tf.float32)))
#
# restore with saved data
ckpt = tf.train.get_checkpoint_state(meta.model_recog_dir)
#
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
#
#
print('begin to train ...')
#
# variables
#
# y_s = self.graph.get_tensor_by_name('y-input/shape:0')
# y_i = self.graph.get_tensor_by_name('y-input/indices:0')
# y_v = self.graph.get_tensor_by_name('y-input/values:0')
#
# <tf.Operation 'y-input/shape' type=Placeholder>,
# <tf.Operation 'y-input/values' type=Placeholder>,
# <tf.Operation 'y-input/indices' type=Placeholder>]
#
#
num_samples = len(data_train['x'])
#
# start training
start_time = time.time()
begin_time = start_time
#
step = sess.run(self.global_step)
#
train_step_half = int(self.train_steps * 0.5)
train_step_quar = int(self.train_steps * 0.75)
#
while step < self.train_steps:
#
if step == train_step_half:
sess.run(tf.assign(self.learning_rate, tf.constant(self.learning_rate_base/10, dtype=tf.float32)))
if step == train_step_quar:
sess.run(tf.assign(self.learning_rate, tf.constant(self.learning_rate_base/100, dtype=tf.float32)))
#
# save and validate
if step % self.valid_freq == 0:
#
print('save model to ckpt ...')
saver.save(sess, os.path.join(meta.model_recog_dir, meta.model_recog_name), \
global_step = step)
#
print('validating ...')
model_v = ModelRecog()
model_v.validate(data_valid, step)
#
#
# train
index_batch = random.sample(range(num_samples), self.batch_size)
#
images= [data_train['x'][i] for i in index_batch]
targets = [data_train['y'][i] for i in index_batch]
#
w_arr = [item.shape[1] for item in images]
max_w = max(w_arr)
img_padd = []
for item in images:
if item.shape[1] != max_w:
img_zeros = np.zeros(shape=[meta.height_norm, max_w - item.shape[1], 3], dtype=np.float32)
item = np.concatenate([item, img_zeros], axis = 1)
img_padd.append(item)
images = img_padd
#
# targets_sparse_value
tsv = model_def.convert2SparseTensorValue(targets)
#
#
#feed_dict = {self.x: images, self.w: w_arr, y_s: tsv.dense_shape, y_i: tsv.indices, y_v: tsv.values}
#
feed_dict = {self.x: images, self.w: w_arr, self.y: tsv}
#
#print(width)
#conv_v = sess.run(conv_feat, feed_dict)
#print(len(conv_v))
# sess.run
_, loss_value, step, lr = sess.run([self.train_op, self.loss, self.global_step, self.learning_rate],\
feed_dict)
#
if step % self.loss_freq == 0:
#
curr_time = time.time()
#
print('step: %d, loss: %g, lr: %g, sect_time: %.1f, total_time: %.1f' %
(step, loss_value, lr, curr_time - begin_time, curr_time - start_time) )
#
begin_time = curr_time
#
#
#
def validate(self, data_valid, step):
#
# valid_result save-path
if not os.path.exists(meta.dir_results_valid): os.mkdir(meta.dir_results_valid)
#
self.create_graph_all(training = False)
#
with self.graph.as_default():
#
saver = tf.train.Saver()
with tf.Session(config = self.sess_config) as sess:
#
tf.global_variables_initializer().run()
#sess.run(tf.assign(self.is_train, tf.constant(False, dtype=tf.bool)))
#
# restore with saved data
ckpt = tf.train.get_checkpoint_state(meta.model_recog_dir)
#
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
#
# pb
constant_graph = graph_util.convert_variables_to_constants(sess, sess.graph_def, output_node_names = \
['rnn_logits/BiasAdd','seq_len',\
'CTCBeamSearchDecoder'])
with tf.gfile.FastGFile(self.pb_file, mode='wb') as f:
f.write(constant_graph.SerializeToString())
#
# variables
#
result_i = self.graph.get_tensor_by_name('CTCBeamSearchDecoder:0')
result_v = self.graph.get_tensor_by_name('CTCBeamSearchDecoder:1')
result_s = self.graph.get_tensor_by_name('CTCBeamSearchDecoder:2')
#
#y_s = self.graph.get_tensor_by_name('y-input/shape:0')
#y_i = self.graph.get_tensor_by_name('y-input/indices:0')
#y_v = self.graph.get_tensor_by_name('y-input/values:0')
#
# <tf.Operation 'y-input/shape' type=Placeholder>,
# <tf.Operation 'y-input/values' type=Placeholder>,
# <tf.Operation 'y-input/indices' type=Placeholder>]
#
# validate
num_samples = len(data_valid['x'])
num_right = 0
num_total = 0
num_batches = np.ceil(num_samples * 1.0 / self.batch_size_valid)
#
curr = 0
batch_start = 0
batch_end = self.batch_size_valid
while batch_start < num_samples:
#
images = data_valid['x'][batch_start:batch_end]
targets = data_valid['y'][batch_start:batch_end]
#
w_arr = [item.shape[1] for item in images]
max_w = max(w_arr)
img_padd = []
for item in images:
if item.shape[1] != max_w:
img_zeros = np.zeros(shape=[meta.height_norm, max_w - item.shape[1], 3], dtype=np.float32)
item = np.concatenate([item, img_zeros], axis = 1)
img_padd.append(item)
images = img_padd
#
batch_start = batch_start + self.batch_size_valid
batch_end = min(batch_end + self.batch_size_valid, num_samples)
#
# targets_sparse_value
tsv = model_def.convert2SparseTensorValue(targets)
#
feed_dict = {self.x: images, self.w: w_arr, self.y: tsv}
#
results, loss, seq_length, d_i, d_v, d_s = \
sess.run([self.result_logits, self.loss, self.seq_len, \
result_i, result_v, result_s], feed_dict)
#
decoded = tf.SparseTensorValue(indices = d_i, values = d_v, dense_shape = d_s)
#
curr += 1
print('curr: %d / %d, loss: %f' % (curr, num_batches, loss))
#
#print(targets)
#print(results)
#print(decoded)
#
trans = model_def.convert2ListLabels(decoded)
#
print(targets)
print(trans)
#
diff = [item[0]-item[1] for item in zip(targets[0], trans[0])]
num_right += diff.count(0)
num_total += sum([ len(word) for word in targets ])
#
#
acc = num_right/num_total
#
print('validation finished, char-accuracy: %f' % acc )
print()
#