inception_v2.py 26 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# 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.
# ==============================================================================
"""Contains the definition for inception v2 classification network."""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

21
22
import tensorflow.compat.v1 as tf
import tf_slim as slim
23

Alex Kurakin's avatar
Alex Kurakin committed
24
25
from nets import inception_utils

26
# pylint: disable=g-long-lambda
27
trunc_normal = lambda stddev: tf.truncated_normal_initializer(
28
    0.0, stddev)
29
30
31
32
33
34


def inception_v2_base(inputs,
                      final_endpoint='Mixed_5c',
                      min_depth=16,
                      depth_multiplier=1.0,
Derek Chow's avatar
Derek Chow committed
35
36
                      use_separable_conv=True,
                      data_format='NHWC',
37
                      include_root_block=True,
38
39
40
41
42
43
44
45
46
47
48
49
50
                      scope=None):
  """Inception v2 (6a2).

  Constructs an Inception v2 network from inputs to the given final endpoint.
  This method can construct the network up to the layer inception(5b) as
  described in http://arxiv.org/abs/1502.03167.

  Args:
    inputs: a tensor of shape [batch_size, height, width, channels].
    final_endpoint: specifies the endpoint to construct the network up to. It
      can be one of ['Conv2d_1a_7x7', 'MaxPool_2a_3x3', 'Conv2d_2b_1x1',
      'Conv2d_2c_3x3', 'MaxPool_3a_3x3', 'Mixed_3b', 'Mixed_3c', 'Mixed_4a',
      'Mixed_4b', 'Mixed_4c', 'Mixed_4d', 'Mixed_4e', 'Mixed_5a', 'Mixed_5b',
51
52
53
      'Mixed_5c']. If include_root_block is False, ['Conv2d_1a_7x7',
      'MaxPool_2a_3x3', 'Conv2d_2b_1x1', 'Conv2d_2c_3x3', 'MaxPool_3a_3x3'] will
      not be available.
54
55
56
57
58
59
60
    min_depth: Minimum depth value (number of channels) for all convolution ops.
      Enforced when depth_multiplier < 1, and not an active constraint when
      depth_multiplier >= 1.
    depth_multiplier: Float multiplier for the depth (number of channels)
      for all convolution ops. The value must be greater than zero. Typical
      usage will be to set this value in (0, 1) to reduce the number of
      parameters or computation cost of the model.
Derek Chow's avatar
Derek Chow committed
61
62
63
    use_separable_conv: Use a separable convolution for the first layer
      Conv2d_1a_7x7. If this is False, use a normal convolution instead.
    data_format: Data format of the activations ('NHWC' or 'NCHW').
64
65
    include_root_block: If True, include the convolution and max-pooling layers
      before the inception modules. If False, excludes those layers.
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
    scope: Optional variable_scope.

  Returns:
    tensor_out: output tensor corresponding to the final_endpoint.
    end_points: a set of activations for external use, for example summaries or
                losses.

  Raises:
    ValueError: if final_endpoint is not set to one of the predefined values,
                or depth_multiplier <= 0
  """

  # end_points will collect relevant activations for external use, for example
  # summaries or losses.
  end_points = {}

  # Used to find thinned depths for each layer.
  if depth_multiplier <= 0:
    raise ValueError('depth_multiplier is not greater than zero.')
  depth = lambda d: max(int(d * depth_multiplier), min_depth)

Derek Chow's avatar
Derek Chow committed
87
88
89
90
91
92
93
94
95
  if data_format != 'NHWC' and data_format != 'NCHW':
    raise ValueError('data_format must be either NHWC or NCHW.')
  if data_format == 'NCHW' and use_separable_conv:
    raise ValueError(
        'separable convolution only supports NHWC layout. NCHW data format can'
        ' only be used when use_separable_conv is False.'
    )

  concat_dim = 3 if data_format == 'NHWC' else 1
96
  with tf.variable_scope(scope, 'InceptionV2', [inputs]):
97
    with slim.arg_scope(
Derek Chow's avatar
Derek Chow committed
98
99
100
101
        [slim.conv2d, slim.max_pool2d, slim.avg_pool2d],
        stride=1,
        padding='SAME',
        data_format=data_format):
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
      net = inputs
      if include_root_block:
        # Note that sizes in the comments below assume an input spatial size of
        # 224x224, however, the inputs can be of any size greater 32x32.

        # 224 x 224 x 3
        end_point = 'Conv2d_1a_7x7'

        if use_separable_conv:
          # depthwise_multiplier here is different from depth_multiplier.
          # depthwise_multiplier determines the output channels of the initial
          # depthwise conv (see docs for tf.nn.separable_conv2d), while
          # depth_multiplier controls the # channels of the subsequent 1x1
          # convolution. Must have
          #   in_channels * depthwise_multipler <= out_channels
          # so that the separable convolution is not overparameterized.
          depthwise_multiplier = min(int(depth(64) / 3), 8)
          net = slim.separable_conv2d(
              inputs,
              depth(64), [7, 7],
              depth_multiplier=depthwise_multiplier,
              stride=2,
              padding='SAME',
              weights_initializer=trunc_normal(1.0),
              scope=end_point)
        else:
          # Use a normal convolution instead of a separable convolution.
          net = slim.conv2d(
              inputs,
              depth(64), [7, 7],
              stride=2,
              weights_initializer=trunc_normal(1.0),
              scope=end_point)
        end_points[end_point] = net
        if end_point == final_endpoint:
          return net, end_points
        # 112 x 112 x 64
        end_point = 'MaxPool_2a_3x3'
        net = slim.max_pool2d(net, [3, 3], scope=end_point, stride=2)
        end_points[end_point] = net
        if end_point == final_endpoint:
          return net, end_points
        # 56 x 56 x 64
        end_point = 'Conv2d_2b_1x1'
Derek Chow's avatar
Derek Chow committed
147
        net = slim.conv2d(
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
            net,
            depth(64), [1, 1],
            scope=end_point,
            weights_initializer=trunc_normal(0.1))
        end_points[end_point] = net
        if end_point == final_endpoint:
          return net, end_points
        # 56 x 56 x 64
        end_point = 'Conv2d_2c_3x3'
        net = slim.conv2d(net, depth(192), [3, 3], scope=end_point)
        end_points[end_point] = net
        if end_point == final_endpoint:
          return net, end_points
        # 56 x 56 x 192
        end_point = 'MaxPool_3a_3x3'
        net = slim.max_pool2d(net, [3, 3], scope=end_point, stride=2)
        end_points[end_point] = net
        if end_point == final_endpoint:
          return net, end_points

168
169
170
      # 28 x 28 x 192
      # Inception module.
      end_point = 'Mixed_3b'
171
172
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
173
          branch_0 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')
174
        with tf.variable_scope('Branch_1'):
175
176
177
178
179
180
          branch_1 = slim.conv2d(
              net, depth(64), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(branch_1, depth(64), [3, 3],
                                 scope='Conv2d_0b_3x3')
181
        with tf.variable_scope('Branch_2'):
182
183
184
185
186
187
188
189
          branch_2 = slim.conv2d(
              net, depth(64), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],
                                 scope='Conv2d_0b_3x3')
          branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],
                                 scope='Conv2d_0c_3x3')
190
        with tf.variable_scope('Branch_3'):
191
192
193
194
195
          branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')
          branch_3 = slim.conv2d(
              branch_3, depth(32), [1, 1],
              weights_initializer=trunc_normal(0.1),
              scope='Conv2d_0b_1x1')
Derek Chow's avatar
Derek Chow committed
196
197
        net = tf.concat(
            axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])
198
199
200
201
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
      # 28 x 28 x 256
      end_point = 'Mixed_3c'
202
203
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
204
          branch_0 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')
205
        with tf.variable_scope('Branch_1'):
206
207
208
209
210
211
          branch_1 = slim.conv2d(
              net, depth(64), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(branch_1, depth(96), [3, 3],
                                 scope='Conv2d_0b_3x3')
212
        with tf.variable_scope('Branch_2'):
213
214
215
216
217
218
219
220
          branch_2 = slim.conv2d(
              net, depth(64), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],
                                 scope='Conv2d_0b_3x3')
          branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],
                                 scope='Conv2d_0c_3x3')
221
        with tf.variable_scope('Branch_3'):
222
223
224
225
226
          branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')
          branch_3 = slim.conv2d(
              branch_3, depth(64), [1, 1],
              weights_initializer=trunc_normal(0.1),
              scope='Conv2d_0b_1x1')
Derek Chow's avatar
Derek Chow committed
227
228
        net = tf.concat(
            axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])
229
230
231
232
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
      # 28 x 28 x 320
      end_point = 'Mixed_4a'
233
234
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
235
236
237
238
239
240
          branch_0 = slim.conv2d(
              net, depth(128), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_0 = slim.conv2d(branch_0, depth(160), [3, 3], stride=2,
                                 scope='Conv2d_1a_3x3')
241
        with tf.variable_scope('Branch_1'):
242
243
244
245
246
247
248
249
          branch_1 = slim.conv2d(
              net, depth(64), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(
              branch_1, depth(96), [3, 3], scope='Conv2d_0b_3x3')
          branch_1 = slim.conv2d(
              branch_1, depth(96), [3, 3], stride=2, scope='Conv2d_1a_3x3')
250
        with tf.variable_scope('Branch_2'):
251
252
          branch_2 = slim.max_pool2d(
              net, [3, 3], stride=2, scope='MaxPool_1a_3x3')
Derek Chow's avatar
Derek Chow committed
253
        net = tf.concat(axis=concat_dim, values=[branch_0, branch_1, branch_2])
254
255
256
257
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
      # 14 x 14 x 576
      end_point = 'Mixed_4b'
258
259
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
260
          branch_0 = slim.conv2d(net, depth(224), [1, 1], scope='Conv2d_0a_1x1')
261
        with tf.variable_scope('Branch_1'):
262
263
264
265
266
267
          branch_1 = slim.conv2d(
              net, depth(64), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(
              branch_1, depth(96), [3, 3], scope='Conv2d_0b_3x3')
268
        with tf.variable_scope('Branch_2'):
269
270
271
272
273
274
275
276
          branch_2 = slim.conv2d(
              net, depth(96), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_2 = slim.conv2d(branch_2, depth(128), [3, 3],
                                 scope='Conv2d_0b_3x3')
          branch_2 = slim.conv2d(branch_2, depth(128), [3, 3],
                                 scope='Conv2d_0c_3x3')
277
        with tf.variable_scope('Branch_3'):
278
279
280
281
282
          branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')
          branch_3 = slim.conv2d(
              branch_3, depth(128), [1, 1],
              weights_initializer=trunc_normal(0.1),
              scope='Conv2d_0b_1x1')
Derek Chow's avatar
Derek Chow committed
283
284
        net = tf.concat(
            axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])
285
286
287
288
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
      # 14 x 14 x 576
      end_point = 'Mixed_4c'
289
290
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
291
          branch_0 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')
292
        with tf.variable_scope('Branch_1'):
293
294
295
296
297
298
          branch_1 = slim.conv2d(
              net, depth(96), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(branch_1, depth(128), [3, 3],
                                 scope='Conv2d_0b_3x3')
299
        with tf.variable_scope('Branch_2'):
300
301
302
303
304
305
306
307
          branch_2 = slim.conv2d(
              net, depth(96), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_2 = slim.conv2d(branch_2, depth(128), [3, 3],
                                 scope='Conv2d_0b_3x3')
          branch_2 = slim.conv2d(branch_2, depth(128), [3, 3],
                                 scope='Conv2d_0c_3x3')
308
        with tf.variable_scope('Branch_3'):
309
310
311
312
313
          branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')
          branch_3 = slim.conv2d(
              branch_3, depth(128), [1, 1],
              weights_initializer=trunc_normal(0.1),
              scope='Conv2d_0b_1x1')
Derek Chow's avatar
Derek Chow committed
314
315
        net = tf.concat(
            axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])
316
317
318
319
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
      # 14 x 14 x 576
      end_point = 'Mixed_4d'
320
321
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
322
          branch_0 = slim.conv2d(net, depth(160), [1, 1], scope='Conv2d_0a_1x1')
323
        with tf.variable_scope('Branch_1'):
324
325
326
327
328
329
          branch_1 = slim.conv2d(
              net, depth(128), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(branch_1, depth(160), [3, 3],
                                 scope='Conv2d_0b_3x3')
330
        with tf.variable_scope('Branch_2'):
331
332
333
334
335
336
337
338
          branch_2 = slim.conv2d(
              net, depth(128), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_2 = slim.conv2d(branch_2, depth(160), [3, 3],
                                 scope='Conv2d_0b_3x3')
          branch_2 = slim.conv2d(branch_2, depth(160), [3, 3],
                                 scope='Conv2d_0c_3x3')
339
        with tf.variable_scope('Branch_3'):
340
341
342
343
344
          branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')
          branch_3 = slim.conv2d(
              branch_3, depth(96), [1, 1],
              weights_initializer=trunc_normal(0.1),
              scope='Conv2d_0b_1x1')
Derek Chow's avatar
Derek Chow committed
345
346
        net = tf.concat(
            axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])
347
348
349
350
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
      # 14 x 14 x 576
      end_point = 'Mixed_4e'
351
352
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
353
          branch_0 = slim.conv2d(net, depth(96), [1, 1], scope='Conv2d_0a_1x1')
354
        with tf.variable_scope('Branch_1'):
355
356
357
358
359
360
          branch_1 = slim.conv2d(
              net, depth(128), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(branch_1, depth(192), [3, 3],
                                 scope='Conv2d_0b_3x3')
361
        with tf.variable_scope('Branch_2'):
362
363
364
365
366
367
368
369
          branch_2 = slim.conv2d(
              net, depth(160), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_2 = slim.conv2d(branch_2, depth(192), [3, 3],
                                 scope='Conv2d_0b_3x3')
          branch_2 = slim.conv2d(branch_2, depth(192), [3, 3],
                                 scope='Conv2d_0c_3x3')
370
        with tf.variable_scope('Branch_3'):
371
372
373
374
375
          branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')
          branch_3 = slim.conv2d(
              branch_3, depth(96), [1, 1],
              weights_initializer=trunc_normal(0.1),
              scope='Conv2d_0b_1x1')
Derek Chow's avatar
Derek Chow committed
376
377
        net = tf.concat(
            axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])
378
379
380
381
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
      # 14 x 14 x 576
      end_point = 'Mixed_5a'
382
383
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
384
385
386
387
388
389
          branch_0 = slim.conv2d(
              net, depth(128), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_0 = slim.conv2d(branch_0, depth(192), [3, 3], stride=2,
                                 scope='Conv2d_1a_3x3')
390
        with tf.variable_scope('Branch_1'):
391
392
393
394
395
396
397
398
          branch_1 = slim.conv2d(
              net, depth(192), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(branch_1, depth(256), [3, 3],
                                 scope='Conv2d_0b_3x3')
          branch_1 = slim.conv2d(branch_1, depth(256), [3, 3], stride=2,
                                 scope='Conv2d_1a_3x3')
399
        with tf.variable_scope('Branch_2'):
400
401
          branch_2 = slim.max_pool2d(net, [3, 3], stride=2,
                                     scope='MaxPool_1a_3x3')
Derek Chow's avatar
Derek Chow committed
402
403
        net = tf.concat(
            axis=concat_dim, values=[branch_0, branch_1, branch_2])
404
405
406
407
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
      # 7 x 7 x 1024
      end_point = 'Mixed_5b'
408
409
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
410
          branch_0 = slim.conv2d(net, depth(352), [1, 1], scope='Conv2d_0a_1x1')
411
        with tf.variable_scope('Branch_1'):
412
413
414
415
416
417
          branch_1 = slim.conv2d(
              net, depth(192), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(branch_1, depth(320), [3, 3],
                                 scope='Conv2d_0b_3x3')
418
        with tf.variable_scope('Branch_2'):
419
420
421
422
423
424
425
426
          branch_2 = slim.conv2d(
              net, depth(160), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_2 = slim.conv2d(branch_2, depth(224), [3, 3],
                                 scope='Conv2d_0b_3x3')
          branch_2 = slim.conv2d(branch_2, depth(224), [3, 3],
                                 scope='Conv2d_0c_3x3')
427
        with tf.variable_scope('Branch_3'):
428
429
430
431
432
          branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')
          branch_3 = slim.conv2d(
              branch_3, depth(128), [1, 1],
              weights_initializer=trunc_normal(0.1),
              scope='Conv2d_0b_1x1')
Derek Chow's avatar
Derek Chow committed
433
434
        net = tf.concat(
            axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])
435
436
437
438
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
      # 7 x 7 x 1024
      end_point = 'Mixed_5c'
439
440
      with tf.variable_scope(end_point):
        with tf.variable_scope('Branch_0'):
441
          branch_0 = slim.conv2d(net, depth(352), [1, 1], scope='Conv2d_0a_1x1')
442
        with tf.variable_scope('Branch_1'):
443
444
445
446
447
448
          branch_1 = slim.conv2d(
              net, depth(192), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_1 = slim.conv2d(branch_1, depth(320), [3, 3],
                                 scope='Conv2d_0b_3x3')
449
        with tf.variable_scope('Branch_2'):
450
451
452
453
454
455
456
457
          branch_2 = slim.conv2d(
              net, depth(192), [1, 1],
              weights_initializer=trunc_normal(0.09),
              scope='Conv2d_0a_1x1')
          branch_2 = slim.conv2d(branch_2, depth(224), [3, 3],
                                 scope='Conv2d_0b_3x3')
          branch_2 = slim.conv2d(branch_2, depth(224), [3, 3],
                                 scope='Conv2d_0c_3x3')
458
        with tf.variable_scope('Branch_3'):
459
460
461
462
463
          branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
          branch_3 = slim.conv2d(
              branch_3, depth(128), [1, 1],
              weights_initializer=trunc_normal(0.1),
              scope='Conv2d_0b_1x1')
Derek Chow's avatar
Derek Chow committed
464
465
        net = tf.concat(
            axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])
466
467
468
469
470
471
472
473
474
475
476
477
478
479
        end_points[end_point] = net
        if end_point == final_endpoint: return net, end_points
    raise ValueError('Unknown final endpoint %s' % final_endpoint)


def inception_v2(inputs,
                 num_classes=1000,
                 is_training=True,
                 dropout_keep_prob=0.8,
                 min_depth=16,
                 depth_multiplier=1.0,
                 prediction_fn=slim.softmax,
                 spatial_squeeze=True,
                 reuse=None,
480
481
                 scope='InceptionV2',
                 global_pool=False):
482
483
484
485
486
487
488
489
490
  """Inception v2 model for classification.

  Constructs an Inception v2 network for classification as described in
  http://arxiv.org/abs/1502.03167.

  The default image size used to train this network is 224x224.

  Args:
    inputs: a tensor of shape [batch_size, height, width, channels].
491
492
493
    num_classes: number of predicted classes. If 0 or None, the logits layer
      is omitted and the input features to the logits layer (before dropout)
      are returned instead.
494
495
496
497
498
499
500
501
502
503
    is_training: whether is training or not.
    dropout_keep_prob: the percentage of activation values that are retained.
    min_depth: Minimum depth value (number of channels) for all convolution ops.
      Enforced when depth_multiplier < 1, and not an active constraint when
      depth_multiplier >= 1.
    depth_multiplier: Float multiplier for the depth (number of channels)
      for all convolution ops. The value must be greater than zero. Typical
      usage will be to set this value in (0, 1) to reduce the number of
      parameters or computation cost of the model.
    prediction_fn: a function to get predictions out of logits.
Derek Chow's avatar
Derek Chow committed
504
505
    spatial_squeeze: if True, logits is of shape [B, C], if false logits is of
        shape [B, 1, 1, C], where B is batch_size and C is number of classes.
506
507
508
    reuse: whether or not the network and its variables should be reused. To be
      able to reuse 'scope' must be given.
    scope: Optional variable_scope.
509
510
511
512
    global_pool: Optional boolean flag to control the avgpooling before the
      logits layer. If false or unset, pooling is done with a fixed window
      that reduces default-sized inputs to 1x1, while larger inputs lead to
      larger outputs. If true, any input size is pooled down to 1x1.
513
514

  Returns:
515
516
517
    net: a Tensor with the logits (pre-softmax activations) if num_classes
      is a non-zero integer, or the non-dropped-out input to the logits layer
      if num_classes is 0 or None.
518
519
520
521
522
523
524
525
526
527
528
    end_points: a dictionary from components of the network to the corresponding
      activation.

  Raises:
    ValueError: if final_endpoint is not set to one of the predefined values,
                or depth_multiplier <= 0
  """
  if depth_multiplier <= 0:
    raise ValueError('depth_multiplier is not greater than zero.')

  # Final pooling and prediction
529
  with tf.variable_scope(
530
      scope, 'InceptionV2', [inputs], reuse=reuse) as scope:
531
532
533
534
535
    with slim.arg_scope([slim.batch_norm, slim.dropout],
                        is_training=is_training):
      net, end_points = inception_v2_base(
          inputs, scope=scope, min_depth=min_depth,
          depth_multiplier=depth_multiplier)
536
      with tf.variable_scope('Logits'):
537
538
        if global_pool:
          # Global average pooling.
539
540
          net = tf.reduce_mean(
              input_tensor=net, axis=[1, 2], keepdims=True, name='global_pool')
541
542
543
544
545
546
547
548
549
          end_points['global_pool'] = net
        else:
          # Pooling with a fixed kernel size.
          kernel_size = _reduced_kernel_size_for_small_input(net, [7, 7])
          net = slim.avg_pool2d(net, kernel_size, padding='VALID',
                                scope='AvgPool_1a_{}x{}'.format(*kernel_size))
          end_points['AvgPool_1a'] = net
        if not num_classes:
          return net, end_points
550
551
        # 1 x 1 x 1024
        net = slim.dropout(net, keep_prob=dropout_keep_prob, scope='Dropout_1b')
552
        end_points['PreLogits'] = net
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
        logits = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,
                             normalizer_fn=None, scope='Conv2d_1c_1x1')
        if spatial_squeeze:
          logits = tf.squeeze(logits, [1, 2], name='SpatialSqueeze')
      end_points['Logits'] = logits
      end_points['Predictions'] = prediction_fn(logits, scope='Predictions')
  return logits, end_points
inception_v2.default_image_size = 224


def _reduced_kernel_size_for_small_input(input_tensor, kernel_size):
  """Define kernel size which is automatically reduced for small input.

  If the shape of the input images is unknown at graph construction time this
  function assumes that the input images are is large enough.

  Args:
    input_tensor: input tensor of size [batch_size, height, width, channels].
    kernel_size: desired kernel size of length 2: [kernel_height, kernel_width]

  Returns:
    a tensor with the kernel size.

  TODO(jrru): Make this function work with unknown shapes. Theoretically, this
  can be done with the code below. Problems are two-fold: (1) If the shape was
  known, it will be lost. (2) inception.slim.ops._two_element_tuple cannot
  handle tensors that define the kernel size.
      shape = tf.shape(input_tensor)
Derek Chow's avatar
Derek Chow committed
581
582
      return = tf.stack([tf.minimum(shape[1], kernel_size[0]),
                         tf.minimum(shape[2], kernel_size[1])])
583
584
585
586
587
588
589
590
591
592
593

  """
  shape = input_tensor.get_shape().as_list()
  if shape[1] is None or shape[2] is None:
    kernel_size_out = kernel_size
  else:
    kernel_size_out = [min(shape[1], kernel_size[0]),
                       min(shape[2], kernel_size[1])]
  return kernel_size_out


Alex Kurakin's avatar
Alex Kurakin committed
594
inception_v2_arg_scope = inception_utils.inception_arg_scope