resnet_v2.py 15.1 KB
Newer Older
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
# 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 definitions for the preactivation form of Residual Networks.

Residual networks (ResNets) were originally proposed in:
[1] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
    Deep Residual Learning for Image Recognition. arXiv:1512.03385

The full preactivation 'v2' ResNet variant implemented in this module was
introduced by:
[2] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
    Identity Mappings in Deep Residual Networks. arXiv: 1603.05027

The key difference of the full preactivation 'v2' variant compared to the
'v1' variant in [1] is the use of batch normalization before every weight layer.

Typical use:

   from tensorflow.contrib.slim.nets import resnet_v2

ResNet-101 for image classification into 1000 classes:

   # inputs has shape [batch, 224, 224, 3]
   with slim.arg_scope(resnet_v2.resnet_arg_scope()):
      net, end_points = resnet_v2.resnet_v2_101(inputs, 1000, is_training=False)

ResNet-101 for semantic segmentation into 21 classes:

   # inputs has shape [batch, 513, 513, 3]
42
   with slim.arg_scope(resnet_v2.resnet_arg_scope()):
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
      net, end_points = resnet_v2.resnet_v2_101(inputs,
                                                21,
                                                is_training=False,
                                                global_pool=False,
                                                output_stride=16)
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import tensorflow as tf

from nets import resnet_utils

slim = tf.contrib.slim
resnet_arg_scope = resnet_utils.resnet_arg_scope


@slim.add_arg_scope
def bottleneck(inputs, depth, depth_bottleneck, stride, rate=1,
               outputs_collections=None, scope=None):
  """Bottleneck residual unit variant with BN before convolutions.

  This is the full preactivation residual unit variant proposed in [2]. See
  Fig. 1(b) of [2] for its definition. Note that we use here the bottleneck
  variant which has an extra bottleneck layer.

  When putting together two consecutive ResNet blocks that use this unit, one
  should use stride = 2 in the last unit of the first block.

  Args:
    inputs: A tensor of size [batch, height, width, channels].
    depth: The depth of the ResNet unit output.
    depth_bottleneck: The depth of the bottleneck layers.
    stride: The ResNet unit's stride. Determines the amount of downsampling of
      the units output compared to its input.
    rate: An integer, rate for atrous convolution.
    outputs_collections: Collection to add the ResNet unit output.
    scope: Optional variable_scope.

  Returns:
    The ResNet unit's output.
  """
  with tf.variable_scope(scope, 'bottleneck_v2', [inputs]) as sc:
    depth_in = slim.utils.last_dimension(inputs.get_shape(), min_rank=4)
    preact = slim.batch_norm(inputs, activation_fn=tf.nn.relu, scope='preact')
    if depth == depth_in:
      shortcut = resnet_utils.subsample(inputs, stride, 'shortcut')
    else:
      shortcut = slim.conv2d(preact, depth, [1, 1], stride=stride,
                             normalizer_fn=None, activation_fn=None,
                             scope='shortcut')

    residual = slim.conv2d(preact, depth_bottleneck, [1, 1], stride=1,
                           scope='conv1')
    residual = resnet_utils.conv2d_same(residual, depth_bottleneck, 3, stride,
                                        rate=rate, scope='conv2')
    residual = slim.conv2d(residual, depth, [1, 1], stride=1,
                           normalizer_fn=None, activation_fn=None,
                           scope='conv3')

    output = shortcut + residual

    return slim.utils.collect_named_outputs(outputs_collections,
107
                                            sc.name,
108
109
110
111
112
113
114
115
116
117
                                            output)


def resnet_v2(inputs,
              blocks,
              num_classes=None,
              is_training=True,
              global_pool=True,
              output_stride=None,
              include_root_block=True,
Derek Chow's avatar
Derek Chow committed
118
              spatial_squeeze=True,
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
              reuse=None,
              scope=None):
  """Generator for v2 (preactivation) ResNet models.

  This function generates a family of ResNet v2 models. See the resnet_v2_*()
  methods for specific model instantiations, obtained by selecting different
  block instantiations that produce ResNets of various depths.

  Training for image classification on Imagenet is usually done with [224, 224]
  inputs, resulting in [7, 7] feature maps at the output of the last ResNet
  block for the ResNets defined in [1] that have nominal stride equal to 32.
  However, for dense prediction tasks we advise that one uses inputs with
  spatial dimensions that are multiples of 32 plus 1, e.g., [321, 321]. In
  this case the feature maps at the ResNet output will have spatial shape
  [(height - 1) / output_stride + 1, (width - 1) / output_stride + 1]
  and corners exactly aligned with the input image corners, which greatly
  facilitates alignment of the features to the image. Using as input [225, 225]
  images results in [8, 8] feature maps at the output of the last ResNet block.

  For dense prediction tasks, the ResNet needs to run in fully-convolutional
  (FCN) mode and global_pool needs to be set to False. The ResNets in [1, 2] all
  have nominal stride equal to 32 and a good choice in FCN mode is to use
  output_stride=16 in order to increase the density of the computed features at
  small computational and memory overhead, cf. http://arxiv.org/abs/1606.00915.

  Args:
    inputs: A tensor of size [batch, height_in, width_in, channels].
    blocks: A list of length equal to the number of ResNet blocks. Each element
      is a resnet_utils.Block object describing the units in the block.
148
149
150
    num_classes: Number of predicted classes for classification tasks.
      If 0 or None, we return the features before the logit layer.
    is_training: whether batch_norm layers are in training mode.
151
152
153
154
155
156
157
158
    global_pool: If True, we perform global average pooling before computing the
      logits. Set to True for image classification, False for dense prediction.
    output_stride: If None, then the output will be computed at the nominal
      network stride. If output_stride is not None, it specifies the requested
      ratio of input to output spatial resolution.
    include_root_block: If True, include the initial convolution followed by
      max-pooling, if False excludes it. If excluded, `inputs` should be the
      results of an activation-less convolution.
Neal Wu's avatar
Neal Wu committed
159
    spatial_squeeze: if True, logits is of shape [B, C], if false logits is
160
        of shape [B, 1, 1, C], where B is batch_size and C is number of classes.
161
162
163
        To use this parameter, the input images must be smaller than 300x300
        pixels, in which case the output logit layer does not contain spatial
        information and can be removed.
164
165
166
167
168
169
170
171
172
    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.


  Returns:
    net: A rank-4 tensor of size [batch, height_out, width_out, channels_out].
      If global_pool is False, then height_out and width_out are reduced by a
      factor of output_stride compared to the respective height_in and width_in,
173
174
175
176
      else both height_out and width_out equal one. If num_classes is 0 or None,
      then net is the output of the last ResNet block, potentially after global
      average pooling. If num_classes is a non-zero integer, net contains the
      pre-softmax activations.
177
178
179
180
181
182
183
    end_points: A dictionary from components of the network to the corresponding
      activation.

  Raises:
    ValueError: If the target output_stride is not valid.
  """
  with tf.variable_scope(scope, 'resnet_v2', [inputs], reuse=reuse) as sc:
184
    end_points_collection = sc.original_name_scope + '_end_points'
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
    with slim.arg_scope([slim.conv2d, bottleneck,
                         resnet_utils.stack_blocks_dense],
                        outputs_collections=end_points_collection):
      with slim.arg_scope([slim.batch_norm], is_training=is_training):
        net = inputs
        if include_root_block:
          if output_stride is not None:
            if output_stride % 4 != 0:
              raise ValueError('The output_stride needs to be a multiple of 4.')
            output_stride /= 4
          # We do not include batch normalization or activation functions in
          # conv1 because the first ResNet unit will perform these. Cf.
          # Appendix of [2].
          with slim.arg_scope([slim.conv2d],
                              activation_fn=None, normalizer_fn=None):
            net = resnet_utils.conv2d_same(net, 64, 7, stride=2, scope='conv1')
          net = slim.max_pool2d(net, [3, 3], stride=2, scope='pool1')
        net = resnet_utils.stack_blocks_dense(net, blocks, output_stride)
        # This is needed because the pre-activation variant does not have batch
        # normalization or activation functions in the residual unit output. See
        # Appendix of [2].
        net = slim.batch_norm(net, activation_fn=tf.nn.relu, scope='postnorm')
207
208
209
210
        # Convert end_points_collection into a dictionary of end_points.
        end_points = slim.utils.convert_collection_to_dict(
            end_points_collection)

211
212
213
        if global_pool:
          # Global average pooling.
          net = tf.reduce_mean(net, [1, 2], name='pool5', keep_dims=True)
214
          end_points['global_pool'] = net
215
216
217
        if num_classes is not None:
          net = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,
                            normalizer_fn=None, scope='logits')
218
          end_points[sc.name + '/logits'] = net
219
220
          if spatial_squeeze:
            net = tf.squeeze(net, [1, 2], name='SpatialSqueeze')
221
            end_points[sc.name + '/spatial_squeeze'] = net
222
223
          end_points['predictions'] = slim.softmax(net, scope='predictions')
        return net, end_points
224
resnet_v2.default_image_size = 224
225
226


derekjchow's avatar
derekjchow committed
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
def resnet_v2_block(scope, base_depth, num_units, stride):
  """Helper function for creating a resnet_v2 bottleneck block.

  Args:
    scope: The scope of the block.
    base_depth: The depth of the bottleneck layer for each unit.
    num_units: The number of units in the block.
    stride: The stride of the block, implemented as a stride in the last unit.
      All other units have stride=1.

  Returns:
    A resnet_v2 bottleneck block.
  """
  return resnet_utils.Block(scope, bottleneck, [{
      'depth': base_depth * 4,
      'depth_bottleneck': base_depth,
      'stride': 1
  }] * (num_units - 1) + [{
      'depth': base_depth * 4,
      'depth_bottleneck': base_depth,
      'stride': stride
  }])
resnet_v2.default_image_size = 224


252
253
254
255
256
def resnet_v2_50(inputs,
                 num_classes=None,
                 is_training=True,
                 global_pool=True,
                 output_stride=None,
Derek Chow's avatar
Derek Chow committed
257
                 spatial_squeeze=True,
258
259
260
261
                 reuse=None,
                 scope='resnet_v2_50'):
  """ResNet-50 model of [1]. See resnet_v2() for arg and return description."""
  blocks = [
derekjchow's avatar
derekjchow committed
262
263
264
265
266
      resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),
      resnet_v2_block('block2', base_depth=128, num_units=4, stride=2),
      resnet_v2_block('block3', base_depth=256, num_units=6, stride=2),
      resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),
  ]
267
268
  return resnet_v2(inputs, blocks, num_classes, is_training=is_training,
                   global_pool=global_pool, output_stride=output_stride,
269
270
                   include_root_block=True, spatial_squeeze=spatial_squeeze,
                   reuse=reuse, scope=scope)
271
272
resnet_v2_50.default_image_size = resnet_v2.default_image_size

273
274
275
276
277
278

def resnet_v2_101(inputs,
                  num_classes=None,
                  is_training=True,
                  global_pool=True,
                  output_stride=None,
Derek Chow's avatar
Derek Chow committed
279
                  spatial_squeeze=True,
280
281
282
283
                  reuse=None,
                  scope='resnet_v2_101'):
  """ResNet-101 model of [1]. See resnet_v2() for arg and return description."""
  blocks = [
derekjchow's avatar
derekjchow committed
284
285
286
287
288
      resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),
      resnet_v2_block('block2', base_depth=128, num_units=4, stride=2),
      resnet_v2_block('block3', base_depth=256, num_units=23, stride=2),
      resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),
  ]
289
290
  return resnet_v2(inputs, blocks, num_classes, is_training=is_training,
                   global_pool=global_pool, output_stride=output_stride,
291
292
                   include_root_block=True, spatial_squeeze=spatial_squeeze,
                   reuse=reuse, scope=scope)
293
resnet_v2_101.default_image_size = resnet_v2.default_image_size
294
295
296
297
298
299
300


def resnet_v2_152(inputs,
                  num_classes=None,
                  is_training=True,
                  global_pool=True,
                  output_stride=None,
Derek Chow's avatar
Derek Chow committed
301
                  spatial_squeeze=True,
302
303
304
305
                  reuse=None,
                  scope='resnet_v2_152'):
  """ResNet-152 model of [1]. See resnet_v2() for arg and return description."""
  blocks = [
derekjchow's avatar
derekjchow committed
306
307
308
309
310
      resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),
      resnet_v2_block('block2', base_depth=128, num_units=8, stride=2),
      resnet_v2_block('block3', base_depth=256, num_units=36, stride=2),
      resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),
  ]
311
312
  return resnet_v2(inputs, blocks, num_classes, is_training=is_training,
                   global_pool=global_pool, output_stride=output_stride,
313
314
                   include_root_block=True, spatial_squeeze=spatial_squeeze,
                   reuse=reuse, scope=scope)
315
resnet_v2_152.default_image_size = resnet_v2.default_image_size
316
317
318
319
320
321
322


def resnet_v2_200(inputs,
                  num_classes=None,
                  is_training=True,
                  global_pool=True,
                  output_stride=None,
Derek Chow's avatar
Derek Chow committed
323
                  spatial_squeeze=True,
324
325
326
327
                  reuse=None,
                  scope='resnet_v2_200'):
  """ResNet-200 model of [2]. See resnet_v2() for arg and return description."""
  blocks = [
derekjchow's avatar
derekjchow committed
328
329
330
331
332
      resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),
      resnet_v2_block('block2', base_depth=128, num_units=24, stride=2),
      resnet_v2_block('block3', base_depth=256, num_units=36, stride=2),
      resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),
  ]
333
334
  return resnet_v2(inputs, blocks, num_classes, is_training=is_training,
                   global_pool=global_pool, output_stride=output_stride,
335
336
                   include_root_block=True, spatial_squeeze=spatial_squeeze,
                   reuse=reuse, scope=scope)
337
resnet_v2_200.default_image_size = resnet_v2.default_image_size