-
Notifications
You must be signed in to change notification settings - Fork 131
/
train.py
executable file
·321 lines (261 loc) · 12.2 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
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
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
#!/usr/bin/env python
"""
Copyright 2017-2018 Fizyr (https://fizyr.com)
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
"""
import argparse
import os
import sys
import keras
import keras_retinanet.losses
from keras_retinanet.callbacks import RedirectModel
from keras_retinanet.utils.config import read_config_file, parse_anchor_parameters
from keras_retinanet.utils.gpu import setup_gpu
from keras_retinanet.utils.keras_version import check_keras_version
from keras_retinanet.utils.model import freeze as freeze_model
from keras_retinanet.utils.transform import random_transform_generator
# Allow relative imports when being executed as script.
if __name__ == "__main__" and __package__ is None:
sys.path.insert(0, os.path.join(os.path.dirname(__file__), '..', '..'))
import keras_maskrcnn.bin
__package__ = "keras_maskrcnn.bin"
# Change these to absolute imports if you copy this script outside the keras_retinanet package.
from .. import losses
from .. import models
from ..callbacks.eval import Evaluate
def model_with_weights(model, weights, skip_mismatch):
if weights is not None:
model.load_weights(weights, by_name=True, skip_mismatch=skip_mismatch)
return model
def create_models(backbone_retinanet, num_classes, weights, freeze_backbone=False, class_specific_filter=True, anchor_params=None):
modifier = freeze_model if freeze_backbone else None
model = model_with_weights(
backbone_retinanet(
num_classes,
nms=True,
class_specific_filter=class_specific_filter,
modifier=modifier,
anchor_params=anchor_params
), weights=weights, skip_mismatch=True)
training_model = model
prediction_model = model
# compile model
training_model.compile(
loss={
'regression' : keras_retinanet.losses.smooth_l1(),
'classification': keras_retinanet.losses.focal(),
'masks' : losses.mask(),
},
optimizer=keras.optimizers.adam(lr=1e-5, clipnorm=0.001)
)
return model, training_model, prediction_model
def create_callbacks(model, training_model, prediction_model, validation_generator, args, create_evaluation=Evaluate):
callbacks = []
# save the prediction model
if args.snapshots:
# ensure directory created first; otherwise h5py will error after epoch.
os.makedirs(args.snapshot_path, exist_ok=True)
checkpoint = keras.callbacks.ModelCheckpoint(
os.path.join(
args.snapshot_path,
'{backbone}_{dataset_type}_{{epoch:02d}}.h5'.format(backbone=args.backbone, dataset_type=args.dataset_type)
),
verbose=1
)
checkpoint = RedirectModel(checkpoint, model)
callbacks.append(checkpoint)
tensorboard_callback = None
if args.tensorboard_dir:
tensorboard_callback = keras.callbacks.TensorBoard(
log_dir = args.tensorboard_dir,
histogram_freq = 0,
batch_size = args.batch_size,
write_graph = True,
write_grads = False,
write_images = False,
embeddings_freq = 0,
embeddings_layer_names = None,
embeddings_metadata = None
)
callbacks.append(tensorboard_callback)
if args.evaluation and validation_generator:
if args.dataset_type == 'coco':
from ..callbacks.coco import CocoEval
# use prediction model for evaluation
evaluation = CocoEval(validation_generator)
elif create_evaluation:
evaluation = create_evaluation(validation_generator, tensorboard=tensorboard_callback, weighted_average=args.weighted_average)
else:
evaluation = Evaluate(validation_generator, tensorboard=tensorboard_callback, weighted_average=args.weighted_average)
evaluation = RedirectModel(evaluation, prediction_model)
callbacks.append(evaluation)
callbacks.append(keras.callbacks.ReduceLROnPlateau(
monitor = 'loss',
factor = 0.1,
patience = 2,
verbose = 1,
mode = 'auto',
epsilon = 0.0001,
cooldown = 0,
min_lr = 0
))
return callbacks
def create_generators(args):
# create random transform generator for augmenting training data
transform_generator = random_transform_generator(flip_x_chance=0.5)
if args.dataset_type == 'coco':
# import here to prevent unnecessary dependency on cocoapi
from ..preprocessing.coco import CocoGenerator
train_generator = CocoGenerator(
args.coco_path,
'train2017',
transform_generator=transform_generator,
batch_size=args.batch_size,
config=args.config
)
validation_generator = None
if args.evaluation:
validation_generator = CocoGenerator(
args.coco_path,
'val2017',
batch_size=args.batch_size,
config=args.config
)
elif args.dataset_type == 'csv':
from ..preprocessing.csv_generator import CSVGenerator
train_generator = CSVGenerator(
args.annotations,
args.classes,
transform_generator=transform_generator,
batch_size=args.batch_size,
config=args.config
)
if args.val_annotations:
validation_generator = CSVGenerator(
args.val_annotations,
args.classes,
batch_size=args.batch_size,
config=args.config
)
else:
validation_generator = None
else:
raise ValueError('Invalid data type received: {}'.format(args.dataset_type))
return train_generator, validation_generator
def check_args(parsed_args):
"""
Function to check for inherent contradictions within parsed arguments.
For example, batch_size < num_gpus
Intended to raise errors prior to backend initialisation.
:param parsed_args: parser.parse_args()
:return: parsed_args
"""
return parsed_args
def parse_args(args):
parser = argparse.ArgumentParser(description='Simple training script for training a RetinaNet mask network.')
subparsers = parser.add_subparsers(help='Arguments for specific dataset types.', dest='dataset_type')
subparsers.required = True
coco_parser = subparsers.add_parser('coco')
coco_parser.add_argument('coco_path', help='Path to dataset directory (ie. /tmp/COCO).')
csv_parser = subparsers.add_parser('csv')
csv_parser.add_argument('annotations', help='Path to CSV file containing annotations for training.')
csv_parser.add_argument('classes', help='Path to a CSV file containing class label mapping.')
csv_parser.add_argument('--val-annotations', help='Path to CSV file containing annotations for validation (optional).')
group = parser.add_mutually_exclusive_group()
group.add_argument('--snapshot', help='Resume training from a snapshot.')
group.add_argument('--imagenet-weights', help='Initialize the model with pretrained imagenet weights. This is the default behaviour.', action='store_const', const=True, default=True)
group.add_argument('--weights', help='Initialize the model with weights from a file.')
group.add_argument('--no-weights', help='Don\'t initialize the model with any weights.', dest='imagenet_weights', action='store_const', const=False)
parser.add_argument('--backbone', help='Backbone model used by retinanet.', default='resnet50', type=str)
parser.add_argument('--batch-size', help='Size of the batches.', default=1, type=int)
parser.add_argument('--gpu', help='Id of the GPU to use (as reported by nvidia-smi).')
parser.add_argument('--epochs', help='Number of epochs to train.', type=int, default=50)
parser.add_argument('--steps', help='Number of steps per epoch.', type=int, default=10000)
parser.add_argument('--snapshot-path', help='Path to store snapshots of models during training (defaults to \'./snapshots\')', default='./snapshots')
parser.add_argument('--tensorboard-dir', help='Log directory for Tensorboard output', default='./logs')
parser.add_argument('--no-snapshots', help='Disable saving snapshots.', dest='snapshots', action='store_false')
parser.add_argument('--no-evaluation', help='Disable per epoch evaluation.', dest='evaluation', action='store_false')
parser.add_argument('--freeze-backbone', help='Freeze training of backbone layers.', action='store_true')
parser.add_argument('--no-class-specific-filter', help='Disables class specific filtering.', dest='class_specific_filter', action='store_false')
parser.add_argument('--config', help='Path to a configuration parameters .ini file.')
parser.add_argument('--weighted-average', help='Compute the mAP using the weighted average of precisions among classes.', action='store_true')
# Fit generator arguments
parser.add_argument('--workers', help='Number of multiprocessing workers. To disable multiprocessing, set workers to 0', type=int, default=1)
parser.add_argument('--max-queue-size', help='Queue length for multiprocessing workers in fit generator.', type=int, default=10)
return check_args(parser.parse_args(args))
def main(args=None):
# parse arguments
if args is None:
args = sys.argv[1:]
args = parse_args(args)
# make sure keras is the minimum required version
check_keras_version()
# create object that stores backbone information
backbone = models.backbone(args.backbone)
# optionally choose specific GPU
if args.gpu:
setup_gpu(args.gpu)
# optionally load config parameters
if args.config:
args.config = read_config_file(args.config)
# create the generators
train_generator, validation_generator = create_generators(args)
# create the model
if args.snapshot is not None:
print('Loading model, this may take a second...')
model = models.load_model(args.snapshot, backbone_name=args.backbone)
training_model = model
prediction_model = model
else:
weights = args.weights
# default to imagenet if nothing else is specified
if weights is None and args.imagenet_weights:
weights = backbone.download_imagenet()
anchor_params = None
if args.config and 'anchor_parameters' in args.config:
anchor_params = parse_anchor_parameters(args.config)
print('Creating model, this may take a second...')
model, training_model, prediction_model = create_models(
backbone_retinanet=backbone.maskrcnn,
num_classes=train_generator.num_classes(),
weights=weights,
freeze_backbone=args.freeze_backbone,
class_specific_filter=args.class_specific_filter,
anchor_params=anchor_params
)
# print model summary
print(model.summary())
# create the callbacks
callbacks = create_callbacks(
model,
training_model,
prediction_model,
validation_generator,
args,
)
# Use multiprocessing if workers > 0
if args.workers > 0:
use_multiprocessing = True
else:
use_multiprocessing = False
# start training
training_model.fit_generator(
generator=train_generator,
steps_per_epoch=args.steps,
epochs=args.epochs,
verbose=1,
callbacks=callbacks,
workers=args.workers,
use_multiprocessing=use_multiprocessing,
max_queue_size=args.max_queue_size
)
if __name__ == '__main__':
main()