-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathaugmentation_training.py
219 lines (176 loc) · 10.4 KB
/
augmentation_training.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
from glob import glob
import keras.applications as ka
from keras.layers import Dense
from keras.optimizers import SGD
from keras.preprocessing import image
from keras.models import Model
from keras.layers import Dense, GlobalAveragePooling2D
from keras import backend as K
from data_generator import MYGenerator
import os
import numpy as np, pandas as pd
import smtplib
import keras.callbacks
from keras.layers import Input
import time, datetime
from keras.models import load_model
from keras.preprocessing.image import ImageDataGenerator
base_models = {
'resnet50': (ka.ResNet50, r'..\keras-pretrained-models\resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5',
ka.resnet50.preprocess_input),
'inception': (ka.InceptionV3, r'..\keras-pretrained-models\inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5',
ka.inception_v3.preprocess_input),
'vgg19': (ka.VGG19, r'..\keras-pretrained-models\resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5',
ka.vgg19.preprocess_input),
'inception_resnet_v2': (ka.InceptionResNetV2, ' ',
ka.inception_resnet_v2.preprocess_input),
'xception': (ka.Xception, r'..\keras-pretrained-models\inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5',
ka.xception.preprocess_input),
'mobilenet_v2': (ka.MobileNetV2, r'..\keras-pretrained-models\resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5',
ka.mobilenet_v2.preprocess_input)
}
class Logger:
def __init__(self, file_name):
self.log_file_name = file_name
def log(self, msg):
with open(self.log_file_name, 'a') as f:
time_str = str(datetime.datetime.today())
f.write('\n\n' + time_str + '\n'+ str(msg))
def save_stats(msg):
with open('stats.txt', 'a') as f:
f.write(str(msg))
if __name__ == '__main__':
logger = Logger('log.txt')
num_classes = 120
image_input_size = (224, 224)
train_epochs = 150
batch_size = 32
seed = 42
classes = pd.read_csv('./data/classes.csv')
classes = classes['class'].tolist()
for base_mode_name in base_models.keys():
if base_mode_name != 'inception_resnet_v2':
continue
base_model_constructor = base_models[base_mode_name][0]
weights_path = 'imagenet' # base_models[base_mode_name][1]
base_model_pre_process = base_models[base_mode_name][2]
input_tensor = Input(shape=(*image_input_size, 3)) # this assumes K.image_data_format() == 'channels_last'
base_model = base_model_constructor(input_tensor=input_tensor,
include_top=False, weights=weights_path,
classes=num_classes, pooling='avg')
x = base_model.output
# let's add a fully-connected layer
x = Dense(1024, activation='relu')(x)
# and a logistic layer -- let's say we have 200 classes
predictions = Dense(num_classes, activation='softmax')(x)
model = Model(inputs=base_model.input, outputs=predictions)
# first: train only the top layers (which were randomly initialized)
# i.e. freeze all convolutional InceptionV3 layers
for layer in base_model.layers[:-10]:
layer.trainable = False
# Learning rate is changed to 0.001
sgd = SGD(lr=1e-3, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd, loss='categorical_crossentropy', metrics=['accuracy'])
train_folder = r'./data_reorganized/train'
train_labels_file = './data/train_0.8_labels.csv'
my_training_batch_generator = ImageDataGenerator(shear_range=0.2,
brightness_range=[0.5, 1.5],
fill_mode='nearest',
zoom_range=[0.9, 1.25],
rotation_range=90,
horizontal_flip=True,
preprocessing_function=base_model_pre_process)\
.flow_from_directory(
directory=train_folder,
target_size=image_input_size,
color_mode="rgb",
batch_size=batch_size,
class_mode="categorical",
shuffle=True,
seed=seed, classes=classes
)
valid_folder = r'./data_reorganized/valid'
valid_labels_file = './data/valid_0.1_labels.csv'
my_validation_batch_generator = ImageDataGenerator(shear_range=0.2,
brightness_range=[0.5, 1.5],
fill_mode='nearest',
zoom_range=[0.9, 1.25],
rotation_range=45,
horizontal_flip=True,
preprocessing_function=base_model_pre_process)\
.flow_from_directory(
directory=valid_folder,
target_size=image_input_size,
color_mode="rgb",
batch_size=batch_size,
class_mode="categorical",
shuffle=True,
seed=seed, classes=classes
)
test_folder = r'./data/test_0.1'
test_labels_file = './data/test_0.1_labels.csv'
my_test_batch_generator = MYGenerator(train_folder=test_folder, labels_file=test_labels_file,
batch_size=batch_size, preprocess_fun_name=base_mode_name)
partially_trained_folder = os.path.join(r'./models_aug/partially_trained_models', base_mode_name)
os.makedirs(partially_trained_folder, exist_ok=True)
save_models_names_format = os.path.join(partially_trained_folder, base_mode_name +
'{epoch:02d} -loss {loss:.3f} -acc{acc:.4f} -'
'val_loss {val_loss: .3f} -val_acc {val_acc:.4f}.hdf5')
check_point_callback = keras.callbacks.ModelCheckpoint\
(save_models_names_format)
num_training_samples = glob(train_folder+'/*/*').__len__()
num_valid_samples = glob(valid_folder+'/*/*').__len__()
logger.log('starting training with base model {:}'.format(base_mode_name))
model.fit_generator(generator=my_training_batch_generator,
validation_data=my_validation_batch_generator,
steps_per_epoch=(num_training_samples // batch_size),
epochs=train_epochs,
verbose=1,
use_multiprocessing=False,
workers=4,
max_queue_size=32, validation_steps=(num_valid_samples//batch_size),
callbacks=[check_point_callback])
logger.log('finished training with {:}'.format(base_mode_name))
# logger.log('weights for model{:} before saving:'.format(base_mode_name))
# logger.log(model.weights)
loss, acc = model.evaluate_generator(generator=my_test_batch_generator,
verbose=1,
steps=1,
use_multiprocessing=True,
workers=16,
max_queue_size=32)
logger.log('model {:} evaluated on test set: acc{:}, loss{:}'.format(base_mode_name, acc, loss))
trained_folder = os.path.join(r'./models_aug/trained_models', base_mode_name)
os.makedirs(trained_folder, exist_ok=True)
os.makedirs(partially_trained_folder, exist_ok=True)
model_save_path = os.path.join(trained_folder, 'trained_{:}_{:}.h5'.format(base_mode_name,
datetime.datetime.today())
.replace(':', '-'))
model.save(model_save_path)
del model
model = load_model(model_save_path)
logger.log('saved model .. now deleting it .. loading and printing weights')
# logger.log(model.weights)
test_folder = r'./data/test_0.1'
test_labels_file = './data/test_0.1_labels.csv'
my_test_batch_generator = MYGenerator(train_folder=test_folder, labels_file=test_labels_file,
batch_size=batch_size, preprocess_fun_name=base_mode_name)
loss, acc = model.evaluate_generator(generator=my_test_batch_generator,
verbose=1,
steps=1,
use_multiprocessing=True,
workers=16,
max_queue_size=32)
logger.log('loaded model {:} evaluated on test set: acc{:}, loss{:}'.format(base_mode_name, acc, loss))
try:
# Gmail Sign In
gmail_sender = '[email protected]'
gmail_passwd = 'Pass4D0GBreed'
server = smtplib.SMTP('smtp.gmail.com', 587)
server.ehlo()
server.starttls()
server.login(gmail_sender, gmail_passwd)
server.sendmail(gmail_sender, ['[email protected]'], 'training has finished')
print('email sent')
except:
print('error sending mail')