-
Notifications
You must be signed in to change notification settings - Fork 88
/
Copy pathtrain.py
178 lines (152 loc) · 6.6 KB
/
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
import gc
import os
import numpy as np
from model_name_encoder import encode_params
from params import args
from transormer import RandomTransformer
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
from datasets.spacenet import generate_ids, MULSpacenetDataset, get_groundtruth
from tools.clr import CyclicLR
from sklearn.model_selection._split import KFold
from keras.callbacks import ModelCheckpoint, EarlyStopping, LearningRateScheduler
from keras.optimizers import RMSprop, Adam, SGD
from losses import make_loss, dice_coef_clipped, binary_crossentropy, dice_coef, ceneterline_loss
from models import make_model
from params import args
import keras.backend as K
def freeze_model(model, freeze_before_layer):
if freeze_before_layer == "ALL":
for l in model.layers:
l.trainable = False
else:
freeze_before_layer_index = -1
for i, l in enumerate(model.layers):
if l.name == freeze_before_layer:
freeze_before_layer_index = i
for l in model.layers[:freeze_before_layer_index]:
l.trainable = False
def main():
if args.crop_size:
print('Using crops of shape ({}, {})'.format(args.crop_size, args.crop_size))
else:
print('Using full size images')
all_ids = np.array(generate_ids(args.data_dirs, args.clahe))
np.random.seed(args.seed)
kfold = KFold(n_splits=args.n_folds, shuffle=True)
splits = [s for s in kfold.split(all_ids)]
folds = [int(f) for f in args.fold.split(",")]
for fold in folds:
encoded_alias = encode_params(args.clahe, args.preprocessing_function, args.stretch_and_mean)
city = "all"
if args.city:
city = args.city.lower()
best_model_file = '{}/{}_{}_{}.h5'.format(args.models_dir, encoded_alias, city, args.network)
channels = 8
if args.ohe_city:
channels = 12
model = make_model(args.network, (None, None, channels))
if args.weights is None:
print('No weights passed, training from scratch')
else:
print('Loading weights from {}'.format(args.weights))
model.load_weights(args.weights, by_name=True)
freeze_model(model, args.freeze_till_layer)
optimizer = RMSprop(lr=args.learning_rate)
if args.optimizer:
if args.optimizer == 'rmsprop':
optimizer = RMSprop(lr=args.learning_rate)
elif args.optimizer == 'adam':
optimizer = Adam(lr=args.learning_rate)
elif args.optimizer == 'sgd':
optimizer = SGD(lr=args.learning_rate, momentum=0.9, nesterov=True)
train_ind, test_ind = splits[fold]
train_ids = all_ids[train_ind]
val_ids = all_ids[test_ind]
if args.city:
val_ids = [id for id in val_ids if args.city in id[0]]
train_ids = [id for id in train_ids if args.city in id[0]]
print('Training fold #{}, {} in train_ids, {} in val_ids'.format(fold, len(train_ids), len(val_ids)))
masks_gt = get_groundtruth(args.data_dirs)
if args.clahe:
template = 'CLAHE-MUL-PanSharpen/MUL-PanSharpen_{id}.tif'
else:
template = 'MUL-PanSharpen/MUL-PanSharpen_{id}.tif'
train_generator = MULSpacenetDataset(
data_dirs=args.data_dirs,
wdata_dir=args.wdata_dir,
clahe=args.clahe,
batch_size=args.batch_size,
image_ids=train_ids,
masks_dict=masks_gt,
image_name_template=template,
seed=args.seed,
ohe_city=args.ohe_city,
stretch_and_mean=args.stretch_and_mean,
preprocessing_function=args.preprocessing_function,
crops_per_image=args.crops_per_image,
crop_shape=(args.crop_size, args.crop_size),
random_transformer=RandomTransformer(horizontal_flip=True, vertical_flip=True),
)
val_generator = MULSpacenetDataset(
data_dirs=args.data_dirs,
wdata_dir=args.wdata_dir,
clahe=args.clahe,
batch_size=1,
image_ids=val_ids,
image_name_template=template,
masks_dict=masks_gt,
seed=args.seed,
ohe_city=args.ohe_city,
stretch_and_mean=args.stretch_and_mean,
preprocessing_function=args.preprocessing_function,
shuffle=False,
crops_per_image=1,
crop_shape=(1280, 1280),
random_transformer=None
)
best_model = ModelCheckpoint(filepath=best_model_file, monitor='val_dice_coef_clipped',
verbose=1,
mode='max',
save_best_only=False,
save_weights_only=True)
model.compile(loss=make_loss(args.loss_function),
optimizer=optimizer,
metrics=[dice_coef, binary_crossentropy, ceneterline_loss, dice_coef_clipped])
def schedule_steps(epoch, steps):
for step in steps:
if step[1] > epoch:
print("Setting learning rate to {}".format(step[0]))
return step[0]
print("Setting learning rate to {}".format(steps[-1][0]))
return steps[-1][0]
callbacks = [best_model, EarlyStopping(patience=20, verbose=1, monitor='val_dice_coef_clipped', mode='max')]
if args.schedule is not None:
steps = [(float(step.split(":")[0]), int(step.split(":")[1])) for step in args.schedule.split(",")]
lrSchedule = LearningRateScheduler(lambda epoch: schedule_steps(epoch, steps))
callbacks.insert(0, lrSchedule)
if args.clr is not None:
clr_params = args.clr.split(',')
base_lr = float(clr_params[0])
max_lr = float(clr_params[1])
step = int(clr_params[2])
mode = clr_params[3]
clr = CyclicLR(base_lr=base_lr, max_lr=max_lr, step_size=step, mode=mode)
callbacks.append(clr)
steps_per_epoch = len(all_ids) / args.batch_size + 1
if args.steps_per_epoch:
steps_per_epoch = args.steps_per_epoch
model.fit_generator(
train_generator,
steps_per_epoch=steps_per_epoch,
epochs=args.epochs,
validation_data=val_generator,
validation_steps=len(val_ids),
callbacks=callbacks,
max_queue_size=30,
verbose=1,
workers=args.num_workers)
del model
K.clear_session()
gc.collect()
if __name__ == '__main__':
main()