Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/keras/src/applications/xception.py: 17%
117 statements
« prev ^ index » next coverage.py v7.4.0, created at 2024-01-03 07:57 +0000
« prev ^ index » next coverage.py v7.4.0, created at 2024-01-03 07:57 +0000
1# Copyright 2016 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# ==============================================================================
16"""Xception V1 model for Keras.
18On ImageNet, this model gets to a top-1 validation accuracy of 0.790
19and a top-5 validation accuracy of 0.945.
21Reference:
22 - [Xception: Deep Learning with Depthwise Separable Convolutions](
23 https://arxiv.org/abs/1610.02357) (CVPR 2017)
24"""
26import tensorflow.compat.v2 as tf
28from keras.src import backend
29from keras.src.applications import imagenet_utils
30from keras.src.engine import training
31from keras.src.layers import VersionAwareLayers
32from keras.src.utils import data_utils
33from keras.src.utils import layer_utils
35# isort: off
36from tensorflow.python.util.tf_export import keras_export
38TF_WEIGHTS_PATH = (
39 "https://storage.googleapis.com/tensorflow/keras-applications/"
40 "xception/xception_weights_tf_dim_ordering_tf_kernels.h5"
41)
42TF_WEIGHTS_PATH_NO_TOP = (
43 "https://storage.googleapis.com/tensorflow/keras-applications/"
44 "xception/xception_weights_tf_dim_ordering_tf_kernels_notop.h5"
45)
47layers = VersionAwareLayers()
50@keras_export(
51 "keras.applications.xception.Xception", "keras.applications.Xception"
52)
53def Xception(
54 include_top=True,
55 weights="imagenet",
56 input_tensor=None,
57 input_shape=None,
58 pooling=None,
59 classes=1000,
60 classifier_activation="softmax",
61):
62 """Instantiates the Xception architecture.
64 Reference:
65 - [Xception: Deep Learning with Depthwise Separable Convolutions](
66 https://arxiv.org/abs/1610.02357) (CVPR 2017)
68 For image classification use cases, see
69 [this page for detailed examples](
70 https://keras.io/api/applications/#usage-examples-for-image-classification-models).
72 For transfer learning use cases, make sure to read the
73 [guide to transfer learning & fine-tuning](
74 https://keras.io/guides/transfer_learning/).
76 The default input image size for this model is 299x299.
78 Note: each Keras Application expects a specific kind of input preprocessing.
79 For Xception, call `tf.keras.applications.xception.preprocess_input` on your
80 inputs before passing them to the model.
81 `xception.preprocess_input` will scale input pixels between -1 and 1.
83 Args:
84 include_top: whether to include the fully-connected
85 layer at the top of the network.
86 weights: one of `None` (random initialization),
87 'imagenet' (pre-training on ImageNet),
88 or the path to the weights file to be loaded.
89 input_tensor: optional Keras tensor
90 (i.e. output of `layers.Input()`)
91 to use as image input for the model.
92 input_shape: optional shape tuple, only to be specified
93 if `include_top` is False (otherwise the input shape
94 has to be `(299, 299, 3)`.
95 It should have exactly 3 inputs channels,
96 and width and height should be no smaller than 71.
97 E.g. `(150, 150, 3)` would be one valid value.
98 pooling: Optional pooling mode for feature extraction
99 when `include_top` is `False`.
100 - `None` means that the output of the model will be
101 the 4D tensor output of the
102 last convolutional block.
103 - `avg` means that global average pooling
104 will be applied to the output of the
105 last convolutional block, and thus
106 the output of the model will be a 2D tensor.
107 - `max` means that global max pooling will
108 be applied.
109 classes: optional number of classes to classify images
110 into, only to be specified if `include_top` is True,
111 and if no `weights` argument is specified.
112 classifier_activation: A `str` or callable. The activation function to use
113 on the "top" layer. Ignored unless `include_top=True`. Set
114 `classifier_activation=None` to return the logits of the "top" layer.
115 When loading pretrained weights, `classifier_activation` can only
116 be `None` or `"softmax"`.
118 Returns:
119 A `keras.Model` instance.
120 """
121 if not (weights in {"imagenet", None} or tf.io.gfile.exists(weights)):
122 raise ValueError(
123 "The `weights` argument should be either "
124 "`None` (random initialization), `imagenet` "
125 "(pre-training on ImageNet), "
126 "or the path to the weights file to be loaded."
127 )
129 if weights == "imagenet" and include_top and classes != 1000:
130 raise ValueError(
131 'If using `weights` as `"imagenet"` with `include_top`'
132 " as true, `classes` should be 1000"
133 )
135 # Determine proper input shape
136 input_shape = imagenet_utils.obtain_input_shape(
137 input_shape,
138 default_size=299,
139 min_size=71,
140 data_format=backend.image_data_format(),
141 require_flatten=include_top,
142 weights=weights,
143 )
145 if input_tensor is None:
146 img_input = layers.Input(shape=input_shape)
147 else:
148 if not backend.is_keras_tensor(input_tensor):
149 img_input = layers.Input(tensor=input_tensor, shape=input_shape)
150 else:
151 img_input = input_tensor
153 channel_axis = 1 if backend.image_data_format() == "channels_first" else -1
155 x = layers.Conv2D(
156 32, (3, 3), strides=(2, 2), use_bias=False, name="block1_conv1"
157 )(img_input)
158 x = layers.BatchNormalization(axis=channel_axis, name="block1_conv1_bn")(x)
159 x = layers.Activation("relu", name="block1_conv1_act")(x)
160 x = layers.Conv2D(64, (3, 3), use_bias=False, name="block1_conv2")(x)
161 x = layers.BatchNormalization(axis=channel_axis, name="block1_conv2_bn")(x)
162 x = layers.Activation("relu", name="block1_conv2_act")(x)
164 residual = layers.Conv2D(
165 128, (1, 1), strides=(2, 2), padding="same", use_bias=False
166 )(x)
167 residual = layers.BatchNormalization(axis=channel_axis)(residual)
169 x = layers.SeparableConv2D(
170 128, (3, 3), padding="same", use_bias=False, name="block2_sepconv1"
171 )(x)
172 x = layers.BatchNormalization(axis=channel_axis, name="block2_sepconv1_bn")(
173 x
174 )
175 x = layers.Activation("relu", name="block2_sepconv2_act")(x)
176 x = layers.SeparableConv2D(
177 128, (3, 3), padding="same", use_bias=False, name="block2_sepconv2"
178 )(x)
179 x = layers.BatchNormalization(axis=channel_axis, name="block2_sepconv2_bn")(
180 x
181 )
183 x = layers.MaxPooling2D(
184 (3, 3), strides=(2, 2), padding="same", name="block2_pool"
185 )(x)
186 x = layers.add([x, residual])
188 residual = layers.Conv2D(
189 256, (1, 1), strides=(2, 2), padding="same", use_bias=False
190 )(x)
191 residual = layers.BatchNormalization(axis=channel_axis)(residual)
193 x = layers.Activation("relu", name="block3_sepconv1_act")(x)
194 x = layers.SeparableConv2D(
195 256, (3, 3), padding="same", use_bias=False, name="block3_sepconv1"
196 )(x)
197 x = layers.BatchNormalization(axis=channel_axis, name="block3_sepconv1_bn")(
198 x
199 )
200 x = layers.Activation("relu", name="block3_sepconv2_act")(x)
201 x = layers.SeparableConv2D(
202 256, (3, 3), padding="same", use_bias=False, name="block3_sepconv2"
203 )(x)
204 x = layers.BatchNormalization(axis=channel_axis, name="block3_sepconv2_bn")(
205 x
206 )
208 x = layers.MaxPooling2D(
209 (3, 3), strides=(2, 2), padding="same", name="block3_pool"
210 )(x)
211 x = layers.add([x, residual])
213 residual = layers.Conv2D(
214 728, (1, 1), strides=(2, 2), padding="same", use_bias=False
215 )(x)
216 residual = layers.BatchNormalization(axis=channel_axis)(residual)
218 x = layers.Activation("relu", name="block4_sepconv1_act")(x)
219 x = layers.SeparableConv2D(
220 728, (3, 3), padding="same", use_bias=False, name="block4_sepconv1"
221 )(x)
222 x = layers.BatchNormalization(axis=channel_axis, name="block4_sepconv1_bn")(
223 x
224 )
225 x = layers.Activation("relu", name="block4_sepconv2_act")(x)
226 x = layers.SeparableConv2D(
227 728, (3, 3), padding="same", use_bias=False, name="block4_sepconv2"
228 )(x)
229 x = layers.BatchNormalization(axis=channel_axis, name="block4_sepconv2_bn")(
230 x
231 )
233 x = layers.MaxPooling2D(
234 (3, 3), strides=(2, 2), padding="same", name="block4_pool"
235 )(x)
236 x = layers.add([x, residual])
238 for i in range(8):
239 residual = x
240 prefix = "block" + str(i + 5)
242 x = layers.Activation("relu", name=prefix + "_sepconv1_act")(x)
243 x = layers.SeparableConv2D(
244 728,
245 (3, 3),
246 padding="same",
247 use_bias=False,
248 name=prefix + "_sepconv1",
249 )(x)
250 x = layers.BatchNormalization(
251 axis=channel_axis, name=prefix + "_sepconv1_bn"
252 )(x)
253 x = layers.Activation("relu", name=prefix + "_sepconv2_act")(x)
254 x = layers.SeparableConv2D(
255 728,
256 (3, 3),
257 padding="same",
258 use_bias=False,
259 name=prefix + "_sepconv2",
260 )(x)
261 x = layers.BatchNormalization(
262 axis=channel_axis, name=prefix + "_sepconv2_bn"
263 )(x)
264 x = layers.Activation("relu", name=prefix + "_sepconv3_act")(x)
265 x = layers.SeparableConv2D(
266 728,
267 (3, 3),
268 padding="same",
269 use_bias=False,
270 name=prefix + "_sepconv3",
271 )(x)
272 x = layers.BatchNormalization(
273 axis=channel_axis, name=prefix + "_sepconv3_bn"
274 )(x)
276 x = layers.add([x, residual])
278 residual = layers.Conv2D(
279 1024, (1, 1), strides=(2, 2), padding="same", use_bias=False
280 )(x)
281 residual = layers.BatchNormalization(axis=channel_axis)(residual)
283 x = layers.Activation("relu", name="block13_sepconv1_act")(x)
284 x = layers.SeparableConv2D(
285 728, (3, 3), padding="same", use_bias=False, name="block13_sepconv1"
286 )(x)
287 x = layers.BatchNormalization(
288 axis=channel_axis, name="block13_sepconv1_bn"
289 )(x)
290 x = layers.Activation("relu", name="block13_sepconv2_act")(x)
291 x = layers.SeparableConv2D(
292 1024, (3, 3), padding="same", use_bias=False, name="block13_sepconv2"
293 )(x)
294 x = layers.BatchNormalization(
295 axis=channel_axis, name="block13_sepconv2_bn"
296 )(x)
298 x = layers.MaxPooling2D(
299 (3, 3), strides=(2, 2), padding="same", name="block13_pool"
300 )(x)
301 x = layers.add([x, residual])
303 x = layers.SeparableConv2D(
304 1536, (3, 3), padding="same", use_bias=False, name="block14_sepconv1"
305 )(x)
306 x = layers.BatchNormalization(
307 axis=channel_axis, name="block14_sepconv1_bn"
308 )(x)
309 x = layers.Activation("relu", name="block14_sepconv1_act")(x)
311 x = layers.SeparableConv2D(
312 2048, (3, 3), padding="same", use_bias=False, name="block14_sepconv2"
313 )(x)
314 x = layers.BatchNormalization(
315 axis=channel_axis, name="block14_sepconv2_bn"
316 )(x)
317 x = layers.Activation("relu", name="block14_sepconv2_act")(x)
319 if include_top:
320 x = layers.GlobalAveragePooling2D(name="avg_pool")(x)
321 imagenet_utils.validate_activation(classifier_activation, weights)
322 x = layers.Dense(
323 classes, activation=classifier_activation, name="predictions"
324 )(x)
325 else:
326 if pooling == "avg":
327 x = layers.GlobalAveragePooling2D()(x)
328 elif pooling == "max":
329 x = layers.GlobalMaxPooling2D()(x)
331 # Ensure that the model takes into account
332 # any potential predecessors of `input_tensor`.
333 if input_tensor is not None:
334 inputs = layer_utils.get_source_inputs(input_tensor)
335 else:
336 inputs = img_input
337 # Create model.
338 model = training.Model(inputs, x, name="xception")
340 # Load weights.
341 if weights == "imagenet":
342 if include_top:
343 weights_path = data_utils.get_file(
344 "xception_weights_tf_dim_ordering_tf_kernels.h5",
345 TF_WEIGHTS_PATH,
346 cache_subdir="models",
347 file_hash="0a58e3b7378bc2990ea3b43d5981f1f6",
348 )
349 else:
350 weights_path = data_utils.get_file(
351 "xception_weights_tf_dim_ordering_tf_kernels_notop.h5",
352 TF_WEIGHTS_PATH_NO_TOP,
353 cache_subdir="models",
354 file_hash="b0042744bf5b25fce3cb969f33bebb97",
355 )
356 model.load_weights(weights_path)
357 elif weights is not None:
358 model.load_weights(weights)
360 return model
363@keras_export("keras.applications.xception.preprocess_input")
364def preprocess_input(x, data_format=None):
365 return imagenet_utils.preprocess_input(
366 x, data_format=data_format, mode="tf"
367 )
370@keras_export("keras.applications.xception.decode_predictions")
371def decode_predictions(preds, top=5):
372 return imagenet_utils.decode_predictions(preds, top=top)
375preprocess_input.__doc__ = imagenet_utils.PREPROCESS_INPUT_DOC.format(
376 mode="",
377 ret=imagenet_utils.PREPROCESS_INPUT_RET_DOC_TF,
378 error=imagenet_utils.PREPROCESS_INPUT_ERROR_DOC,
379)
380decode_predictions.__doc__ = imagenet_utils.decode_predictions.__doc__