• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
2#
3# Licensed under the Apache License, Version 2.0 (the "License");
4# you may not use this file except in compliance with the License.
5# You may obtain a copy of the License at
6#
7#     http://www.apache.org/licenses/LICENSE-2.0
8#
9# Unless required by applicable law or agreed to in writing, software
10# distributed under the License is distributed on an "AS IS" BASIS,
11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12# See the License for the specific language governing permissions and
13# limitations under the License.
14# ==============================================================================
15# pylint: disable=invalid-name
16"""ResNet v2 models for Keras.
17
18Reference:
19  - [Identity Mappings in Deep Residual Networks]
20    (https://arxiv.org/abs/1603.05027) (CVPR 2016)
21"""
22
23from tensorflow.python.keras.applications import imagenet_utils
24from tensorflow.python.keras.applications import resnet
25from tensorflow.python.util.tf_export import keras_export
26
27
28@keras_export('keras.applications.resnet_v2.ResNet50V2',
29              'keras.applications.ResNet50V2')
30def ResNet50V2(
31    include_top=True,
32    weights='imagenet',
33    input_tensor=None,
34    input_shape=None,
35    pooling=None,
36    classes=1000,
37    classifier_activation='softmax'):
38  """Instantiates the ResNet50V2 architecture."""
39  def stack_fn(x):
40    x = resnet.stack2(x, 64, 3, name='conv2')
41    x = resnet.stack2(x, 128, 4, name='conv3')
42    x = resnet.stack2(x, 256, 6, name='conv4')
43    return resnet.stack2(x, 512, 3, stride1=1, name='conv5')
44
45  return resnet.ResNet(
46      stack_fn,
47      True,
48      True,
49      'resnet50v2',
50      include_top,
51      weights,
52      input_tensor,
53      input_shape,
54      pooling,
55      classes,
56      classifier_activation=classifier_activation)
57
58
59@keras_export('keras.applications.resnet_v2.ResNet101V2',
60              'keras.applications.ResNet101V2')
61def ResNet101V2(
62    include_top=True,
63    weights='imagenet',
64    input_tensor=None,
65    input_shape=None,
66    pooling=None,
67    classes=1000,
68    classifier_activation='softmax'):
69  """Instantiates the ResNet101V2 architecture."""
70  def stack_fn(x):
71    x = resnet.stack2(x, 64, 3, name='conv2')
72    x = resnet.stack2(x, 128, 4, name='conv3')
73    x = resnet.stack2(x, 256, 23, name='conv4')
74    return resnet.stack2(x, 512, 3, stride1=1, name='conv5')
75
76  return resnet.ResNet(
77      stack_fn,
78      True,
79      True,
80      'resnet101v2',
81      include_top,
82      weights,
83      input_tensor,
84      input_shape,
85      pooling,
86      classes,
87      classifier_activation=classifier_activation)
88
89
90@keras_export('keras.applications.resnet_v2.ResNet152V2',
91              'keras.applications.ResNet152V2')
92def ResNet152V2(
93    include_top=True,
94    weights='imagenet',
95    input_tensor=None,
96    input_shape=None,
97    pooling=None,
98    classes=1000,
99    classifier_activation='softmax'):
100  """Instantiates the ResNet152V2 architecture."""
101  def stack_fn(x):
102    x = resnet.stack2(x, 64, 3, name='conv2')
103    x = resnet.stack2(x, 128, 8, name='conv3')
104    x = resnet.stack2(x, 256, 36, name='conv4')
105    return resnet.stack2(x, 512, 3, stride1=1, name='conv5')
106
107  return resnet.ResNet(
108      stack_fn,
109      True,
110      True,
111      'resnet152v2',
112      include_top,
113      weights,
114      input_tensor,
115      input_shape,
116      pooling,
117      classes,
118      classifier_activation=classifier_activation)
119
120
121@keras_export('keras.applications.resnet_v2.preprocess_input')
122def preprocess_input(x, data_format=None):
123  return imagenet_utils.preprocess_input(
124      x, data_format=data_format, mode='tf')
125
126
127@keras_export('keras.applications.resnet_v2.decode_predictions')
128def decode_predictions(preds, top=5):
129  return imagenet_utils.decode_predictions(preds, top=top)
130
131
132preprocess_input.__doc__ = imagenet_utils.PREPROCESS_INPUT_DOC.format(
133    mode='',
134    ret=imagenet_utils.PREPROCESS_INPUT_RET_DOC_TF,
135    error=imagenet_utils.PREPROCESS_INPUT_ERROR_DOC)
136decode_predictions.__doc__ = imagenet_utils.decode_predictions.__doc__
137
138DOC = """
139
140  Reference:
141  - [Identity Mappings in Deep Residual Networks]
142    (https://arxiv.org/abs/1603.05027) (CVPR 2016)
143
144  For image classification use cases, see
145  [this page for detailed examples](
146    https://keras.io/api/applications/#usage-examples-for-image-classification-models).
147
148  For transfer learning use cases, make sure to read the
149  [guide to transfer learning & fine-tuning](
150    https://keras.io/guides/transfer_learning/).
151
152  Note: each Keras Application expects a specific kind of input preprocessing.
153  For ResNetV2, call `tf.keras.applications.resnet_v2.preprocess_input` on your
154  inputs before passing them to the model.
155  `resnet_v2.preprocess_input` will scale input pixels between -1 and 1.
156
157  Args:
158    include_top: whether to include the fully-connected
159      layer at the top of the network.
160    weights: one of `None` (random initialization),
161      'imagenet' (pre-training on ImageNet),
162      or the path to the weights file to be loaded.
163    input_tensor: optional Keras tensor (i.e. output of `layers.Input()`)
164      to use as image input for the model.
165    input_shape: optional shape tuple, only to be specified
166      if `include_top` is False (otherwise the input shape
167      has to be `(224, 224, 3)` (with `'channels_last'` data format)
168      or `(3, 224, 224)` (with `'channels_first'` data format).
169      It should have exactly 3 inputs channels,
170      and width and height should be no smaller than 32.
171      E.g. `(200, 200, 3)` would be one valid value.
172    pooling: Optional pooling mode for feature extraction
173      when `include_top` is `False`.
174      - `None` means that the output of the model will be
175          the 4D tensor output of the
176          last convolutional block.
177      - `avg` means that global average pooling
178          will be applied to the output of the
179          last convolutional block, and thus
180          the output of the model will be a 2D tensor.
181      - `max` means that global max pooling will
182          be applied.
183    classes: optional number of classes to classify images
184      into, only to be specified if `include_top` is True, and
185      if no `weights` argument is specified.
186    classifier_activation: A `str` or callable. The activation function to use
187      on the "top" layer. Ignored unless `include_top=True`. Set
188      `classifier_activation=None` to return the logits of the "top" layer.
189      When loading pretrained weights, `classifier_activation` can only
190      be `None` or `"softmax"`.
191
192  Returns:
193    A `keras.Model` instance.
194"""
195
196setattr(ResNet50V2, '__doc__', ResNet50V2.__doc__ + DOC)
197setattr(ResNet101V2, '__doc__', ResNet101V2.__doc__ + DOC)
198setattr(ResNet152V2, '__doc__', ResNet152V2.__doc__ + DOC)
199