-
Notifications
You must be signed in to change notification settings - Fork 7
/
facerec_keras_train.py
216 lines (179 loc) · 8.58 KB
/
facerec_keras_train.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
from __future__ import print_function
import argparse
import sys
import os.path
import os
from os import listdir
from os.path import isfile, join
import datetime
#os.environ["CUDA_VISIBLE_DEVICES"] = ""
from keras.preprocessing.image import ImageDataGenerator
from keras.applications import mobilenet,mobilenet_v2,densenet,inception_resnet_v2,inception_v3
from keras.layers import Flatten, Dense, Dropout,GlobalAveragePooling2D, Reshape, Conv2D, Activation, MaxPooling2D, BatchNormalization, Lambda
from keras.models import Model, Sequential,model_from_json
from keras.optimizers import Adam, Nadam
from keras.callbacks import TensorBoard,ModelCheckpoint,EarlyStopping
from keras.applications.imagenet_utils import preprocess_input
from keras.regularizers import l2
from keras.losses import *
import tensorflow as tf
import keras.backend.tensorflow_backend as KTF
from keras import backend as K
import numpy as np
np.random.seed(123)
activ_func,loss_func,class_mode,metric,monitor='softmax',categorical_crossentropy,'categorical','accuracy','val_acc'
input_height=input_width=224 #192 #128 #192 #224 #200
def count_of_files(dir):
#print 'dir=',dir
dirs = next(os.walk(dir))[1]
file_count=0
for d in dirs:
#file_count+=len(next(os.walk(os.path.join(dir,d)))[2])
directory=os.path.join(dir,d)
file_count+=sum(1 for entry in listdir(directory) if isfile(join(directory,entry)))
return file_count,len(dirs)
#net_model=mobilenet_v2
net_model=mobilenet
def get_net():
if False:
return net_model.MobileNetV2(alpha=1.4,include_top=False, weights='imagenet',input_shape=(input_height, input_width, 3))
else:
return net_model.MobileNet(include_top=False, weights='imagenet',input_shape=(input_height, input_width, 3))
def get_predictive_model(base_model,classes_num):
last_model_layer = base_model.output
x = GlobalAveragePooling2D(name='reshape_1')(last_model_layer)
preds = Dense(classes_num, activation=activ_func, kernel_regularizer=l2(4e-5), name='predictions')(x)
f_model = Model(base_model.input, preds)
return f_model
def model_architecture(classes_num):
#net_model = VGG16(weights='imagenet', include_top=False,
# input_shape=(input_height, input_width, 3))
#net_model = ResNet50(weights='imagenet', include_top=False,
# input_shape=(input_height, input_width, 3))
# net_model = Xception(weights='imagenet', include_top=False, input_shape=(input_height, input_width, 3))
# net_model = VGG19(weights='imagenet', include_top=False,
# input_shape=(input_height, input_width, 3))
base_model = get_net()
return get_predictive_model(base_model,classes_num),base_model
def freeze_session(session, keep_var_names=None, output_names=None, clear_devices=True):
from tensorflow.python.framework.graph_util import convert_variables_to_constants
graph = session.graph
with graph.as_default():
freeze_var_names = list(set(v.op.name for v in tf.global_variables()).difference(keep_var_names or []))
output_names = output_names or []
output_names += [v.op.name for v in tf.global_variables()]
input_graph_def = graph.as_graph_def()
if clear_devices:
for node in input_graph_def.node:
node.device = ''
frozen_graph = convert_variables_to_constants(session, input_graph_def,
output_names, freeze_var_names)
return frozen_graph
def load_model(modelpath):
base_model = get_net()
last_model_layer = base_model.output
out = GlobalAveragePooling2D(name='reshape_1')(last_model_layer)
cnn_model = Model(base_model.input, out)
cnn_model.load_weights(modelpath)
return cnn_model,base_model
from tensorflow.python.framework.graph_util import convert_variables_to_constants
def convert_to_tf(modelpath):
K.set_learning_phase(0)
config = tf.ConfigProto(device_count={'CPU':1,'GPU':0})
KTF.set_session(tf.Session(config=config))
filename,ext=os.path.splitext(os.path.basename(modelpath))
print (filename, ext)
if ext=='.hdf5':
model,_ = model_architecture(9131)
model.summary()
model.load_weights(modelpath)
out = model.get_layer('reshape_1').output
print('out=',out)
cnn_model = Model(model.input, out)
#cnn_model.save('models/'+filename+'.h5')
cnn_model.save_weights('weights.h5')
model_json = cnn_model.to_json()
with open("model.json", "w") as json_file:
json_file.write(model_json)
json_file = open('model.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
loaded_model = model_from_json(loaded_model_json)
loaded_model.load_weights('weights.h5')
loaded_model.save('models/'+filename+'.h5')
else:
saved_model = load_model(modelpath)[0]
output_names=[saved_model.output.op.name]
print (output_names)
session=K.get_session()
graph = session.graph
keep_var_names=None
clear_devices=True
with graph.as_default():
freeze_var_names = list(set(v.op.name for v in tf.global_variables()).difference(keep_var_names or []))
output_names = output_names or []
output_names += [v.op.name for v in tf.global_variables()]
input_graph_def = graph.as_graph_def()
if clear_devices:
for node in input_graph_def.node:
node.device = ''
frozen_graph = convert_variables_to_constants(session, input_graph_def,
output_names, freeze_var_names)
tf.train.write_graph(frozen_graph, 'models', filename+'.pb', as_text=False)
if __name__ == '__main__':
#config = tf.ConfigProto()
#config.gpu_options.allocator_type = 'BFC'
#KTF.set_session(tf.Session(config=config))
parser = argparse.ArgumentParser(description="Training script for face recognition.")
parser.add_argument("--datapath", type=str, default='/home/datasets/images/vgg_face_dataset/vggface-2/tf1', help="Path where image data is stored.")
parser.add_argument("--modelpath", type=str, default='', help="Path where trained model is stored.")
train_batch_size=val_batch_size= 32 #76 # 88 #104 #128
args = parser.parse_args()
if args.modelpath!='':
convert_to_tf(args.modelpath)
sys.exit()
train_path=args.datapath + '/train/'
val_path=args.datapath + '/val/'
train_datagen = ImageDataGenerator(shear_range=0.3, #0.2
rotation_range=10,
zoom_range=0.2, #0.1
width_shift_range=0.1,height_shift_range=0.1,
horizontal_flip=True,preprocessing_function=preprocess_input)
test_datagen = ImageDataGenerator(preprocessing_function=preprocess_input)
target_size=(input_height, input_width)
test_generator = test_datagen.flow_from_directory(
val_path,
target_size=target_size,
batch_size=val_batch_size, class_mode=class_mode)
train_generator = train_datagen.flow_from_directory(
train_path,
target_size=target_size,
batch_size=train_batch_size, class_mode=class_mode)
nb_train_samples=train_generator.samples
classes_num=train_generator.num_classes
nb_validation_samples=test_generator.samples
print('after read',nb_train_samples,nb_validation_samples,classes_num)
model_file=''
model,base_model = model_architecture(classes_num)
opt=Adam(lr=1e-3,decay=1e-5)
model.compile(loss=loss_func, optimizer=opt, metrics=[metric])
model.summary()
#filepath="models/casia_mobilenet128"
filepath="models/vgg2_mobilenet2_224"
#filepath="models/gender_mobilenet192_tf15"
#filepath='models/age_mobilenet192_tf1_regr'
#filepath='models/age_simple_tf2'
#saved_weights_filepath=filepath+'-03-0.64.hdf5'
#model.load_weights(saved_weights_filepath)
last_path='-{epoch:02d}-{val_acc:.2f}.hdf5'
checkpoint = ModelCheckpoint(filepath+last_path, monitor=monitor, verbose=1, save_best_only=True, mode='auto')
es=EarlyStopping(monitor='val_acc',patience=2)
callbacks=[checkpoint,es]
history = model.fit_generator(
train_generator,
steps_per_epoch=nb_train_samples // train_batch_size,
epochs=16,
validation_data=test_generator,
validation_steps=nb_validation_samples // val_batch_size,
callbacks=callbacks)
model.save(filepath+'.hdf5')