-
Notifications
You must be signed in to change notification settings - Fork 126
/
dm_resnet.py
620 lines (543 loc) · 27 KB
/
dm_resnet.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
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
# This code is originally from: https://github.com/raghakot/keras-resnet
# Modified by Li Shen for DM challenge.
from keras.models import Model
from keras.layers import (
Input,
Activation,
Dropout,
Dense,
Flatten
)
from keras.layers.merge import concatenate, add
from keras.layers.convolutional import Conv2D
from keras.layers.pooling import (
MaxPooling2D,
AveragePooling2D,
GlobalAveragePooling2D
)
from keras.layers.normalization import BatchNormalization
from keras.layers.core import activations
from keras.regularizers import l1, l2, l1_l2
from keras import backend as K
# import warnings
# warnings.filterwarnings('error')
if K.image_data_format() == 'channels_last':
ROW_AXIS = 1
COL_AXIS = 2
CHANNEL_AXIS = 3
else:
CHANNEL_AXIS = 1
ROW_AXIS = 2
COL_AXIS = 3
# Helper to build a conv -> BN -> relu block
def _conv_bn_relu(nb_filter, nb_row, nb_col, strides=(1, 1),
weight_decay=.0001, dropout=.0, last_block=False):
def f(input):
conv = Conv2D(filters=nb_filter, kernel_size=(nb_row, nb_col),
strides=strides, kernel_initializer="he_normal",
padding="same", kernel_regularizer=l2(weight_decay))(input)
norm = BatchNormalization(axis=CHANNEL_AXIS)(conv)
if last_block:
return norm
else:
relu = Activation("relu")(norm)
return Dropout(dropout)(relu)
return f
# Helper to build a BN -> relu -> conv block
# This is an improved scheme proposed in http://arxiv.org/pdf/1603.05027v2.pdf
def _bn_relu_conv(nb_filter, nb_row, nb_col, strides=(1, 1),
weight_decay=.0001, dropout=.0):
def f(input):
norm = BatchNormalization(axis=CHANNEL_AXIS)(input)
activation = Activation("relu")(norm)
activation = Dropout(dropout)(activation)
return Conv2D(filters=nb_filter, kernel_size=(nb_row, nb_col),
strides=strides, kernel_initializer="he_normal",
padding="same",
kernel_regularizer=l2(weight_decay))(activation)
return f
# Adds a shortcut between input and residual block and merges them with "sum"
def _shortcut(input, residual, weight_decay=.0001, dropout=.0, identity=True,
strides=(1, 1), with_bn=False, org=False):
# Expand channels of shortcut to match residual.
# Stride appropriately to match residual (width, height)
# Should be int if network architecture is correctly configured.
# !!! The dropout argument is just a place holder.
# !!! It shall not be applied to identity mapping.
# stride_width = input._keras_shape[ROW_AXIS] // residual._keras_shape[ROW_AXIS]
# stride_height = input._keras_shape[COL_AXIS] // residual._keras_shape[COL_AXIS]
# equal_channels = residual._keras_shape[CHANNEL_AXIS] == input._keras_shape[CHANNEL_AXIS]
shortcut = input
# 1 X 1 conv if shape is different. Else identity.
# if stride_width > 1 or stride_height > 1 or not equal_channels:
if not identity:
shortcut = Conv2D(filters=residual._keras_shape[CHANNEL_AXIS],
kernel_size=(1, 1), strides=strides,
kernel_initializer="he_normal", padding="valid",
kernel_regularizer=l2(weight_decay))(input)
if with_bn:
shortcut = BatchNormalization(axis=CHANNEL_AXIS)(shortcut)
addition = add([shortcut, residual])
if not org:
return addition
else:
relu = Activation("relu")(addition)
return Dropout(dropout)(relu)
# Builds a residual block with repeating bottleneck blocks.
def _residual_block(block_function, nb_filters, repetitions,
is_first_layer=False, shortcut_with_bn=False,
bottleneck_enlarge_factor=4, **kw_args):
def f(input):
for i in range(repetitions):
init_strides = (1, 1)
identity = True
if i == 0 and not is_first_layer:
init_strides = (2, 2)
if i == 0:
identity = False
input = block_function(nb_filters=nb_filters,
init_strides=init_strides,
identity=identity,
shortcut_with_bn=shortcut_with_bn,
enlarge_factor=bottleneck_enlarge_factor,
**kw_args)(input)
return input
return f
# Basic 3 X 3 convolution blocks.
# Use for resnet with layers <= 34
# Follows improved proposed scheme in http://arxiv.org/pdf/1603.05027v2.pdf
def basic_block(nb_filters, init_strides=(1, 1), identity=True,
shortcut_with_bn=False, enlarge_factor=None, **kw_args):
def f(input):
conv1 = _bn_relu_conv(nb_filters, 3, 3, strides=init_strides, **kw_args)(input)
residual = _bn_relu_conv(nb_filters, 3, 3, **kw_args)(conv1)
return _shortcut(input, residual, identity=identity,
strides=init_strides,
with_bn=shortcut_with_bn, **kw_args)
return f
def basic_block_org(nb_filters, init_strides=(1, 1), identity=True,
shortcut_with_bn=False, enlarge_factor=None, **kw_args):
def f(input):
conv1 = _conv_bn_relu(nb_filters, 3, 3, strides=init_strides, **kw_args)(input)
residual = _conv_bn_relu(nb_filters, 3, 3, last_block=True, **kw_args)(conv1)
return _shortcut(input, residual, identity=identity,
strides=init_strides,
with_bn=shortcut_with_bn, org=True, **kw_args)
return f
# Bottleneck architecture for > 34 layer resnet.
# Follows improved proposed scheme in http://arxiv.org/pdf/1603.05027v2.pdf
# Returns a final conv layer of nb_filters * 4
def bottleneck(nb_filters, init_strides=(1, 1), identity=True,
shortcut_with_bn=False, enlarge_factor=4, **kw_args):
def f(input):
conv_1_1 = _bn_relu_conv(nb_filters, 1, 1, strides=init_strides, **kw_args)(input)
conv_3_3 = _bn_relu_conv(nb_filters, 3, 3, **kw_args)(conv_1_1)
residual = _bn_relu_conv(nb_filters * enlarge_factor, 1, 1, **kw_args)(conv_3_3)
return _shortcut(input, residual, identity=identity,
strides=init_strides,
with_bn=shortcut_with_bn, **kw_args)
return f
def bottleneck_org(nb_filters, init_strides=(1, 1), identity=True,
shortcut_with_bn=False, enlarge_factor=4, **kw_args):
def f(input):
conv_1_1 = _conv_bn_relu(nb_filters, 1, 1, strides=init_strides, **kw_args)(input)
conv_3_3 = _conv_bn_relu(nb_filters, 3, 3, **kw_args)(conv_1_1)
residual = _conv_bn_relu(nb_filters * enlarge_factor, 1, 1,
last_block=True, **kw_args)(conv_3_3)
return _shortcut(input, residual, identity=identity,
strides=init_strides,
with_bn=shortcut_with_bn, org=True, **kw_args)
return f
def _vgg_block(nb_filters, repetitions, dropout=.0, weight_decay=.01):
def f(input):
for i in range(repetitions):
input = Conv2D(nb_filters, (3, 3), padding='same',
kernel_initializer="he_normal",
kernel_regularizer=l2(weight_decay))(input)
input = BatchNormalization()(input)
input = Activation('relu')(input)
input = Dropout(dropout)(input)
input = MaxPooling2D((2, 2), strides=(2, 2))(input)
return input
return f
def add_top_layers(model, image_size, patch_net='resnet50', block_type='resnet',
depths=[512,512], repetitions=[1,1],
block_fn=bottleneck_org, nb_class=2,
shortcut_with_bn=True, bottleneck_enlarge_factor=4,
dropout=.0, weight_decay=.0001,
add_heatmap=False, avg_pool_size=(7,7), return_heatmap=False,
add_conv=True, add_shortcut=False,
hm_strides=(1,1), hm_pool_size=(5,5),
fc_init_units=64, fc_layers=2):
def add_residual_blocks(block):
for depth,repetition in zip(depths, repetitions):
block = _residual_block(
block_fn, depth, repetition,
dropout=dropout, weight_decay=weight_decay,
shortcut_with_bn=shortcut_with_bn,
bottleneck_enlarge_factor=bottleneck_enlarge_factor)(block)
pool = GlobalAveragePooling2D()(block)
dropped = Dropout(dropout)(pool)
return dropped
def add_vgg_blocks(block):
for depth,repetition in zip(depths, repetitions):
block = _vgg_block(depth, repetition,
dropout=dropout,
weight_decay=weight_decay)(block)
pool = GlobalAveragePooling2D()(block)
dropped = Dropout(dropout)(pool)
return dropped
def add_fc_layers(block):
flattened = Flatten()(block)
dropped = Dropout(dropout)(flattened)
units=fc_init_units
for i in xrange(fc_layers):
fc = Dense(units, kernel_initializer="he_normal",
kernel_regularizer=l2(weight_decay))(dropped)
norm = BatchNormalization()(fc)
relu = Activation('relu')(norm)
dropped = Dropout(dropout)(relu)
units /= 2
return dropped, flattened
if patch_net == 'resnet50':
last_kept_layer = model.layers[-5]
elif patch_net == 'yaroslav':
last_kept_layer = model.layers[-3]
else:
last_kept_layer = model.layers[-4]
block = last_kept_layer.output
channels = 1 if patch_net == 'yaroslav' else 3
image_input = Input(shape=(image_size[0], image_size[1], channels))
model0 = Model(inputs=model.inputs, outputs=block)
block = model0(image_input)
if add_heatmap or return_heatmap: # add softmax heatmap.
pool1 = AveragePooling2D(pool_size=avg_pool_size,
strides=hm_strides)(block)
if return_heatmap:
dropped = pool1
else:
dropped = Dropout(dropout)(pool1)
clf_layer = model.layers[-1]
clf_weights = clf_layer.get_weights()
clf_classes = clf_layer.output_shape[1]
if return_heatmap:
activation = activations.softmax(x, axis=CHANNEL_AXIS)
else:
activation = 'relu'
heatmap_layer = Dense(clf_classes, activation=activation,
kernel_regularizer=l2(weight_decay))
heatmap = heatmap_layer(dropped)
heatmap_layer.set_weights(clf_weights)
if return_heatmap:
model_heatmap = Model(inputs=image_input, outputs=heatmap)
return model_heatmap
block = MaxPooling2D(pool_size=hm_pool_size)(heatmap)
top_layer_nb = 8
else:
top_layer_nb = 2
if add_conv:
if block_type == 'resnet':
block = add_residual_blocks(block)
elif block_type == 'vgg':
block = add_vgg_blocks(block)
else:
raise Exception('Unsupported block type: ' + block_type)
else:
block, flattened = add_fc_layers(block)
if add_shortcut and not add_conv:
dense = Dense(nb_class, kernel_initializer="he_normal",
kernel_regularizer=l2(weight_decay))(block)
shortcut = Dense(nb_class, kernel_initializer="he_normal",
kernel_regularizer=l2(weight_decay))(flattened)
addition = add([dense, shortcut])
dense = Activation('softmax')(addition)
else:
dense = Dense(nb_class, kernel_initializer="he_normal",
activation='softmax',
kernel_regularizer=l2(weight_decay))(block)
model_addtop = Model(inputs=image_input, outputs=dense)
# import pdb; pdb.set_trace()
return model_addtop, top_layer_nb
class ResNetBuilder(object):
@staticmethod
def _shared_conv_layers(input_shape, block_fn, repetitions, nb_init_filter=64,
init_filter_size=7, init_conv_stride=2, pool_size=3,
pool_stride=2,
weight_decay=.0001, inp_dropout=.0, hidden_dropout=.0,
shortcut_with_bn=False,
bottleneck_enlarge_factor=4):
'''Create shared conv layers for all inputs
Args:
pool_size ([int]): set to 0 or False to turn off the first max pooling.
'''
if len(input_shape) != 3:
raise Exception("Input shape should be a tuple (nb_channels, nb_rows, nb_cols)")
# Permute dimension order if necessary
if K.image_data_format() == 'channels_last':
input_shape = (input_shape[1], input_shape[2], input_shape[0])
input_ = Input(shape=input_shape)
dropped = Dropout(inp_dropout)(input_)
conv1 = _conv_bn_relu(nb_filter=nb_init_filter,
nb_row=init_filter_size,
nb_col=init_filter_size,
strides=(init_conv_stride, init_conv_stride),
weight_decay=weight_decay, dropout=hidden_dropout)(dropped)
if pool_size:
pool1 = MaxPooling2D(pool_size=(pool_size, pool_size),
strides=(pool_stride, pool_stride),
padding="same")(conv1)
block = pool1
else:
block = conv1
nb_filters = nb_init_filter
for i, r in enumerate(repetitions):
block = _residual_block(
block_fn, nb_filters=nb_filters, repetitions=r,
is_first_layer=(i == 0),
shortcut_with_bn=shortcut_with_bn,
bottleneck_enlarge_factor=bottleneck_enlarge_factor,
weight_decay=weight_decay,
dropout=hidden_dropout)(block)
nb_filters *= 2
# Classifier block
pool2 = GlobalAveragePooling2D()(block)
return input_, pool2
@staticmethod
def l1l2_penalty_reg(alpha=1.0, l1_ratio=0.5):
'''Calculate L1 and L2 penalties for a Keras layer
This follows the same formulation as in the R package glmnet and Sklearn
Args:
alpha ([float]): amount of regularization.
l1_ratio ([float]): portion of L1 penalty. Setting to 1.0 equals
Lasso.
'''
if l1_ratio == .0:
return l2(alpha)
elif l1_ratio == 1.:
return l1(alpha)
else:
return l1_l2(l1_ratio*alpha, 1./2*(1 - l1_ratio)*alpha)
@staticmethod
def build(input_shape, num_outputs, block_fn, repetitions, nb_init_filter=64,
init_filter_size=7, init_conv_stride=2, pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0, shortcut_with_bn=False):
"""
Builds a custom ResNet like architecture.
:param input_shape: The input shape in the form (nb_channels, nb_rows, nb_cols)
:param num_outputs: The number of outputs at final softmax layer
:param block_fn: The block function to use. This is either :func:`basic_block` or :func:`bottleneck`.
The original paper used basic_block for layers < 50
:param repetitions: Number of repetitions of various block units.
At each block unit, the number of filters are doubled and the input size is halved
:return: The keras model.
"""
inputs, flatten_out = ResNetBuilder._shared_conv_layers(
input_shape, block_fn, repetitions,
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay,
inp_dropout=inp_dropout, hidden_dropout=hidden_dropout,
shortcut_with_bn=shortcut_with_bn)
enet_penalty = ResNetBuilder.l1l2_penalty_reg(alpha, l1_ratio)
activation = "softmax" if num_outputs > 1 else "sigmoid"
dense = Dense(units=num_outputs, kernel_initializer="he_normal",
activation=activation, kernel_regularizer=enet_penalty)(flatten_out)
model = Model(inputs=inputs, outputs=dense)
return model
@classmethod
def build_resnet_18(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0):
return cls.build(
input_shape, num_outputs, basic_block, [2, 2, 2, 2],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout)
@classmethod
def build_resnet_34(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0):
return cls.build(
input_shape, num_outputs, basic_block, [3, 4, 6, 3],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout)
@classmethod
def build_resnet_38(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0):
return cls.build(
input_shape, num_outputs, bottleneck, [3, 6, 3],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout)
@classmethod
def build_resnet_50(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0,
shortcut_with_bn=False):
return cls.build(
input_shape, num_outputs, bottleneck, [3, 4, 6, 3],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout, shortcut_with_bn=shortcut_with_bn)
@classmethod
def build_resnet_50_org(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0,
shortcut_with_bn=False):
return cls.build(
input_shape, num_outputs, bottleneck_org, [3, 4, 6, 3],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout, shortcut_with_bn=shortcut_with_bn)
@classmethod
def build_resnet_101(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0):
return cls.build(
input_shape, num_outputs, bottleneck, [3, 4, 23, 3],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout)
@classmethod
def build_resnet_152(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0):
return cls.build(
input_shape, num_outputs, bottleneck, [3, 8, 36, 3],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout)
@classmethod
def build_dm_resnet_14(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0):
return cls.build(
input_shape, num_outputs, bottleneck, [1, 1, 1, 1],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout)
@classmethod
def build_dm_resnet_47rb5(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0):
return cls.build(
input_shape, num_outputs, bottleneck, [3, 3, 3, 3, 3],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout)
@classmethod
def build_dm_resnet_56rb6(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0):
return cls.build(
input_shape, num_outputs, bottleneck, [3, 3, 3, 3, 3, 3],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout)
@classmethod
def build_dm_resnet_65rb7(cls, input_shape, num_outputs,
nb_init_filter=64, init_filter_size=7, init_conv_stride=2,
pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0):
return cls.build(
input_shape, num_outputs, bottleneck, [3, 3, 3, 3, 3, 3, 3],
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay, inp_dropout=inp_dropout,
hidden_dropout=hidden_dropout)
class MultiViewResNetBuilder(ResNetBuilder):
'''Residual net with two inputs
'''
@staticmethod
def build(input_shape, num_outputs, block_fn, repetitions, nb_init_filter=64,
init_filter_size=7, init_conv_stride=2, pool_size=3, pool_stride=2,
weight_decay=.0001, alpha=1., l1_ratio=.5,
inp_dropout=.0, hidden_dropout=.0, shortcut_with_bn=False):
"""
Builds a custom ResNet like architecture.
:param input_shape: Shall be the input shapes for both CC and MLO views.
:param num_outputs: The number of outputs at final softmax layer
:param block_fn: The block function to use. This is either :func:`basic_block` or :func:`bottleneck`.
The original paper used basic_block for layers < 50
:param repetitions: Number of repetitions of various block units.
At each block unit, the number of filters are doubled and the input size is halved
:return: The keras model.
"""
# First, define a shared CNN model for both CC and MLO views.
input_cc, flatten_cc = ResNetBuilder._shared_conv_layers(
input_shape, block_fn, repetitions,
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay,
inp_dropout=inp_dropout, hidden_dropout=hidden_dropout,
shortcut_with_bn=shortcut_with_bn)
input_mlo, flatten_mlo = ResNetBuilder._shared_conv_layers(
input_shape, block_fn, repetitions,
nb_init_filter=nb_init_filter, init_filter_size=init_filter_size,
init_conv_stride=init_conv_stride,
pool_size=pool_size, pool_stride=pool_stride,
weight_decay=weight_decay,
inp_dropout=inp_dropout, hidden_dropout=hidden_dropout,
shortcut_with_bn=shortcut_with_bn)
# Then merge the conv representations of the two views.
merged_repr = concatenate([flatten_cc, flatten_mlo])
enet_penalty = ResNetBuilder.l1l2_penalty_reg(alpha, l1_ratio)
activation = "softmax" if num_outputs > 1 else "sigmoid"
dense = Dense(units=num_outputs, kernel_initializer="he_normal",
activation=activation, kernel_regularizer=enet_penalty)(merged_repr)
discr_model = Model(inputs=[input_cc, input_mlo], outputs=dense)
return discr_model
def main():
model = MultiViewResNetBuilder.build_resnet_50(
(1, 288, 224), 1, inp_dropout=.2, hidden_dropout=.5)
model.compile(loss="binary_crossentropy", optimizer="sgd")
# model.summary()
if __name__ == '__main__':
main()