-
Notifications
You must be signed in to change notification settings - Fork 0
/
training.py
executable file
·209 lines (179 loc) · 9.15 KB
/
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
from keras.utils import multi_gpu_model
from keras.applications import resnet50
from keras.callbacks import CSVLogger, ModelCheckpoint, ReduceLROnPlateau
import tensorflow as tf
import numpy as np
def prepare_model(create_network, input_shape, loss, optimizer, metrics, prior, GPUs=1, L1=0, L2=0, dropout=0, kernel_num=64):
if GPUs==1:
template_model = create_network(input_shape, prior, L1=L1, L2=L2, dropout=dropout, KERNEL_NUM=kernel_num)
model = template_model
elif GPUs>1:
with tf.device('/cpu:0'):
template_model = create_network(input_shape, prior, L1=L1, L2=L2, dropout=dropout, KERNEL_NUM=kernel_num)
model = multi_gpu_model(template_model, gpus=GPUs)
else:
raise ValueError('GPUs needs to be an integer greater than or equal to 1, not {}'.format(GPUs))
template_model.summary()
model.compile(loss=loss, optimizer=optimizer, metrics=metrics)
return (template_model, model)
def data_generator(batch_size, imgs, pos_cubes, neg_cubes):
total_num = imgs.shape[0]
half_batch = batch_size//2
shuffled_indeces = np.arange(total_num)
while True:
np.random.shuffle(shuffled_indeces)
for i in range(total_num//half_batch):
current_indeces = shuffled_indeces[i*half_batch:(i+1)*half_batch]
current_images = imgs[current_indeces]
batch_images = np.concatenate([current_images, current_images], axis=0)
batch_cubes = np.zeros([2*current_indeces.shape[0],8,8,8])
for j,cur_ind in enumerate(current_indeces):
pz,py,px = pos_cubes[cur_ind]
nz,ny,nx = neg_cubes[cur_ind]
batch_cubes[j, px,py,pz] = 1
batch_cubes[half_batch+j, nx,ny,nz] = 0.2
yield batch_images, batch_cubes
def load_data(train_set, val_set):
print('\tLoading Targets')
train_pos_cubes = np.load('Data/for_training/compact/training_positive_cubes.npy').astype(np.int32)[:, 1:]
train_neg_cubes = np.load('Data/for_training/compact/training_negative_cubes.npy').astype(np.int32)[:, 1:]
val_pos_cubes = np.load('Data/for_training/compact/validation_positive_cubes.npy').astype(np.int32)[:, 1:]
val_neg_cubes = np.load('Data/for_training/compact/validation_negative_cubes.npy').astype(np.int32)[:, 1:]
print('\tLoading Images')
#train_images = np.zeros([31062,224,224,3], dtype=np.int8)
if train_set=='small':
train_images = np.load('Data/for_training/compact/small_image_set.npy')
train_pos_cubes = train_pos_cubes[:512]
train_neg_cubes = train_neg_cubes[:512]
elif train_set=='full':
train_images = np.load('Data/for_training/compact/training_image_data.npy')
else:
raise ValueError('{} is not a valid training_set'.format(train_set))
val_images = np.load('Data/for_training/compact/validation_image_data.npy')
if val_set=='random':
val_images = val_images[571:]
val_pos_cubes = val_pos_cubes[571:]
val_neg_cubes = val_neg_cubes[571:]
total_num = 1000-571
elif val_set=='object':
val_images = val_images[:571]
val_pos_cubes = val_pos_cubes[:571]
val_neg_cubes = val_neg_cubes[:571]
total_num = 571
elif val_set=='full':
total_num = 1000
else:
raise ValueError('{} is not a valid validation_set'.format(val_set))
val_images = np.concatenate([val_images,val_images],axis=0)
print('\tPreprocessing Images')
train_images = resnet50.preprocess_input(train_images)
val_images = resnet50.preprocess_input(val_images)
print('\tProcessing Validation Set')
validation_cubes = np.zeros([total_num*2, 8,8,8])
for i in range(total_num):
pz,py,px = val_pos_cubes[i]
validation_cubes[i, px,py,pz] = 1
nz,ny,nx = val_neg_cubes[i]
validation_cubes[i+total_num, nx,ny,nz] = 0.2
return (train_images, val_images), (train_pos_cubes, train_neg_cubes), validation_cubes
def load_network(arch):
if arch == 'resnet_style':
from Models.resnet_style import create_network
elif arch == 'resnet_style_smaller':
from Models.resnet_style_smaller import create_network
elif arch == 'resnet_style_smaller_split':
from Models.resnet_style_smaller_split import create_network
elif arch == 'end_to_end':
from Models.end_to_end import create_network
elif arch == 'end_to_end_sep':
from Models.end_to_end_sep import create_network
elif arch == 'simplest':
from Models.simplest import create_network
elif arch == 'simplest_2':
from Models.simplest_2 import create_network
elif arch == 'resnet_style_smaller_split_no_prior':
from Models.resnet_style_smaller_split_no_prior import create_network
elif arch == 'simple_smaller_split_no_prior':
from Models.simple_smaller_split_no_prior import create_network
elif arch == 'no_reduction_no_prior':
from Models.no_reduction_no_prior import create_network
else:
raise ValueError('{} is not a valid architecture'.format(arch))
return create_network
if __name__ == '__main__':
import configparser
config = configparser.ConfigParser()
config.read('training_config.ini')
# Architecture
arch = config['architecture']['architecture']
kernel_num = config['architecture'].getint('kernel_num')
create_network = load_network(arch)
# Optimizer
opt_name = config['optimizer']['optimizer']
learning_rate = config['optimizer'].getfloat('learning_rate')
clipnorm = config['optimizer'].getfloat('clipnorm')
momentum = config['optimizer'].getfloat('momentum')
if opt_name == 'adam':
from keras.optimizers import Adam
opt = Adam(lr=learning_rate, clipnorm=clipnorm)
elif opt_name == 'sgd':
from keras.optimizers import SGD
opt = SGD(lr=learning_rate, clipnorm=clipnorm, momentum=momentum)
else:
raise ValueError('{} is not a valid optimizer'.format(opt_name))
# Regularizer
L1_REGULARIZER = config['regularizer'].getfloat('l1_penalty')
L2_REGULARIZER = config['regularizer'].getfloat('l2_penalty')
DROPOUT_RATE = config['regularizer'].getfloat('dropout_rate')
# Callbacks
callbacks = []
if config['callbacks'].getboolean('csv_logger'):
csv_fname = config['callbacks']['csv_fname']
csv_logger = CSVLogger(csv_fname)
callbacks.append(csv_logger)
if config['callbacks'].getboolean('model_checkpoint'):
model_fname = config['callbacks']['model_fname']
save_best = config['callbacks'].getboolean('model_save_best')
model_checkpoint = ModelCheckpoint(model_fname, save_best_only=save_best)
callbacks.append(model_checkpoint)
if config['callbacks'].getboolean('reduce_lr_on_plateau'):
monitor = config['callbacks']['reduce_lr_monitor']
factor = config['callbacks']['reduce_lr_factor']
patience = config['callbacks'].getfloat('reduce_lr_patience')
min_delta = config['callbacks'].getfloat('reduce_lr_min_delta')
reduce_lr_on_plateau = ReduceLROnPlateau(monitor=monitor,
factor=factor,
patience=patience,
min_delta=min_delta)
callbacks.append(reduce_lr_on_plateau)
# Dataset names
train_set = config['training']['training_set']
val_set = config['training']['validation_set']
# Hyperparameters
BATCH_SIZE = config['training'].getint('batch_size')
val_set = config['training']['validation_set']
train_set = config['training']['training_set']
print('Building Network')
from Models.loss_and_metrics import single_accuracy, all_way_binary_cross_entropy
(template, model) = prepare_model(create_network, input_shape=[224,224,3],
loss=all_way_binary_cross_entropy,
optimizer=opt,
metrics=[single_accuracy],
GPUs=1,
L1=L1_REGULARIZER,
L2=L2_REGULARIZER,
dropout=DROPOUT_RATE,
prior=np.load('Data/for_training/prior_compact.npy'),
kernel_num=kernel_num)
print('Loading Data')
images, cubes, validation_cubes = load_data(train_set, val_set)
train_images, val_images = images
train_pos_cubes, train_neg_cubes = cubes
print('Configuring Generator')
datagen = data_generator(BATCH_SIZE, train_images, train_pos_cubes, train_neg_cubes)
print('Beginning Training')
batch_per_epoch = (2*train_images.shape[0])/BATCH_SIZE
model.fit_generator(datagen, steps_per_epoch=batch_per_epoch,
epochs=1000,
validation_data=(val_images,validation_cubes),
callbacks=[csv_logger, model_checkpoint])#, reduce_lr_on_plateau])