forked from PaperCodeReview/MoCo-TF
/
main.py
168 lines (134 loc) · 5.57 KB
/
main.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
import os
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
from common import set_seed
from common import get_logger
from common import get_session
from common import get_arguments
from common import search_same
from common import create_stamp
from dataloader import set_dataset
from dataloader import DataLoader
from model import MoCo
from model import set_lincls
from callback import OptionalLearningRateSchedule
from callback import create_callbacks
import tensorflow as tf
def train_moco(args, logger, initial_epoch, strategy, num_workers):
##########################
# Dataset
##########################
trainset = set_dataset(args.task, args.data_path)
steps_per_epoch = args.steps or len(trainset) // args.batch_size
logger.info("TOTAL STEPS OF DATASET FOR TRAINING")
logger.info("========== TRAINSET ==========")
logger.info(f" --> {len(trainset)}")
logger.info(f" --> {steps_per_epoch}")
##########################
# Model & Generator
##########################
with strategy.scope():
model = MoCo(args, logger)
lr_scheduler = OptionalLearningRateSchedule(args, steps_per_epoch, initial_epoch)
model.compile(
optimizer=tf.keras.optimizers.SGD(lr_scheduler, momentum=.9),
loss=tf.keras.losses.sparse_categorical_crossentropy,
metrics=[tf.keras.metrics.TopKCategoricalAccuracy(1, 'acc1', dtype=tf.float32),
tf.keras.metrics.TopKCategoricalAccuracy(5, 'acc5', dtype=tf.float32)],
num_workers=num_workers,
run_eagerly=True)
train_generator = DataLoader(args, 'train', trainset, args.batch_size, num_workers).dataloader
##########################
# Train
##########################
callbacks, initial_epoch = create_callbacks(args, logger, initial_epoch)
if callbacks == -1:
logger.info('Check your model.')
return
elif callbacks == -2:
return
model.fit(
train_generator,
epochs=args.epochs,
callbacks=callbacks,
initial_epoch=initial_epoch,
steps_per_epoch=steps_per_epoch,)
def train_lincls(args, logger, initial_epoch, strategy, num_workers):
assert args.snapshot is not None, 'pretrained weight is needed!'
##########################
# Dataset
##########################
trainset, valset = set_dataset(args.task, args.data_path)
steps_per_epoch = args.steps or len(trainset) // args.batch_size
validation_steps = len(valset) // args.batch_size
logger.info("TOTAL STEPS OF DATASET FOR TRAINING")
logger.info("========== TRAINSET ==========")
logger.info(f" --> {len(trainset)}")
logger.info(f" --> {steps_per_epoch}")
logger.info("=========== VALSET ===========")
logger.info(f" --> {len(valset)}")
logger.info(f" --> {validation_steps}")
##########################
# Model & Generator
##########################
with strategy.scope():
backbone = MoCo(args, logger)
model = set_lincls(args, backbone.encoder_q)
lr_scheduler = OptionalLearningRateSchedule(args, steps_per_epoch, initial_epoch)
model.compile(
optimizer=tf.keras.optimizers.SGD(lr_scheduler, momentum=.9),
metrics=[tf.keras.metrics.TopKCategoricalAccuracy(1, 'acc1', dtype=tf.float32),
tf.keras.metrics.TopKCategoricalAccuracy(5, 'acc5', dtype=tf.float32)],
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True, name='loss'))
train_generator = DataLoader(args, 'train', trainset, args.batch_size, num_workers).dataloader
val_generator = DataLoader(args, 'val', valset, args.batch_size, num_workers).dataloader
##########################
# Train
##########################
callbacks, initial_epoch = create_callbacks(args, logger, initial_epoch)
if callbacks == -1:
logger.info('Check your model.')
return
elif callbacks == -2:
return
model.fit(
train_generator,
validation_data=val_generator,
epochs=args.epochs,
callbacks=callbacks,
initial_epoch=initial_epoch,
steps_per_epoch=steps_per_epoch,
validation_steps=validation_steps)
def main():
set_seed()
args = get_arguments()
args, initial_epoch = search_same(args)
if initial_epoch == -1:
# training was already finished!
return
elif initial_epoch == 0:
# first training or training with snapshot
args.stamp = create_stamp()
get_session(args)
logger = get_logger("MyLogger")
for k, v in vars(args).items():
logger.info("{} : {}".format(k, v))
##########################
# Strategy
##########################
if len(args.gpus.split(',')) > 1:
strategy = tf.distribute.experimental.CentralStorageStrategy()
else:
strategy = tf.distribute.OneDeviceStrategy(device="/gpu:0")
num_workers = strategy.num_replicas_in_sync
assert args.batch_size % num_workers == 0
logger.info('{} : {}'.format(strategy.__class__.__name__, num_workers))
logger.info("GLOBAL BATCH SIZE : {}".format(args.batch_size))
##########################
# Training
##########################
if args.task in ['v1', 'v2']:
train_moco(args, logger, initial_epoch, strategy, num_workers)
else:
train_lincls(args, logger, initial_epoch, strategy, num_workers)
if __name__ == "__main__":
main()