1 # Copyright 2015 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 """Tests for tf.layers.pooling.""" 16 17 from __future__ import absolute_import 18 from __future__ import division 19 from __future__ import print_function 20 21 from tensorflow.python.layers import pooling as pooling_layers 22 from tensorflow.python.ops import array_ops 23 from tensorflow.python.ops import random_ops 24 from tensorflow.python.platform import test 25 26 27 class PoolingTest(test.TestCase): 28 29 def testInvalidDataFormat(self): 30 height, width = 7, 9 31 images = random_ops.random_uniform((5, height, width, 3), seed=1) 32 with self.assertRaisesRegexp(ValueError, 'data_format'): 33 pooling_layers.max_pooling2d(images, 3, strides=2, data_format='invalid') 34 35 def testInvalidStrides(self): 36 height, width = 7, 9 37 images = random_ops.random_uniform((5, height, width, 3), seed=1) 38 with self.assertRaisesRegexp(ValueError, 'strides'): 39 pooling_layers.max_pooling2d(images, 3, strides=(1, 2, 3)) 40 41 with self.assertRaisesRegexp(ValueError, 'strides'): 42 pooling_layers.max_pooling2d(images, 3, strides=None) 43 44 def testInvalidPoolSize(self): 45 height, width = 7, 9 46 images = random_ops.random_uniform((5, height, width, 3), seed=1) 47 with self.assertRaisesRegexp(ValueError, 'pool_size'): 48 pooling_layers.max_pooling2d(images, (1, 2, 3), strides=2) 49 50 with self.assertRaisesRegexp(ValueError, 'pool_size'): 51 pooling_layers.max_pooling2d(images, None, strides=2) 52 53 def testCreateMaxPooling2D(self): 54 height, width = 7, 9 55 images = random_ops.random_uniform((5, height, width, 4)) 56 layer = pooling_layers.MaxPooling2D([2, 2], strides=2) 57 output = layer.apply(images) 58 self.assertListEqual(output.get_shape().as_list(), [5, 3, 4, 4]) 59 60 def testCreateAveragePooling2D(self): 61 height, width = 7, 9 62 images = random_ops.random_uniform((5, height, width, 4)) 63 layer = pooling_layers.AveragePooling2D([2, 2], strides=2) 64 output = layer.apply(images) 65 self.assertListEqual(output.get_shape().as_list(), [5, 3, 4, 4]) 66 67 def testCreateMaxPooling2DChannelsFirst(self): 68 height, width = 7, 9 69 images = random_ops.random_uniform((5, 2, height, width)) 70 layer = pooling_layers.MaxPooling2D([2, 2], 71 strides=1, 72 data_format='channels_first') 73 output = layer.apply(images) 74 self.assertListEqual(output.get_shape().as_list(), [5, 2, 6, 8]) 75 76 def testCreateAveragePooling2DChannelsFirst(self): 77 height, width = 5, 6 78 images = random_ops.random_uniform((3, 4, height, width)) 79 layer = pooling_layers.AveragePooling2D((2, 2), 80 strides=(1, 1), 81 padding='valid', 82 data_format='channels_first') 83 output = layer.apply(images) 84 self.assertListEqual(output.get_shape().as_list(), [3, 4, 4, 5]) 85 86 def testCreateAveragePooling2DChannelsFirstWithNoneBatch(self): 87 height, width = 5, 6 88 images = array_ops.placeholder(dtype='float32', 89 shape=(None, 4, height, width)) 90 layer = pooling_layers.AveragePooling2D((2, 2), 91 strides=(1, 1), 92 padding='valid', 93 data_format='channels_first') 94 output = layer.apply(images) 95 self.assertListEqual(output.get_shape().as_list(), [None, 4, 4, 5]) 96 97 def testCreateMaxPooling1D(self): 98 width = 7 99 channels = 3 100 images = random_ops.random_uniform((5, width, channels)) 101 layer = pooling_layers.MaxPooling1D(2, strides=2) 102 output = layer.apply(images) 103 self.assertListEqual(output.get_shape().as_list(), 104 [5, width // 2, channels]) 105 106 def testCreateAveragePooling1D(self): 107 width = 7 108 channels = 3 109 images = random_ops.random_uniform((5, width, channels)) 110 layer = pooling_layers.AveragePooling1D(2, strides=2) 111 output = layer.apply(images) 112 self.assertListEqual(output.get_shape().as_list(), 113 [5, width // 2, channels]) 114 115 def testCreateMaxPooling1DChannelsFirst(self): 116 width = 7 117 channels = 3 118 images = random_ops.random_uniform((5, channels, width)) 119 layer = pooling_layers.MaxPooling1D( 120 2, strides=2, data_format='channels_first') 121 output = layer.apply(images) 122 self.assertListEqual(output.get_shape().as_list(), 123 [5, channels, width // 2]) 124 125 def testCreateAveragePooling1DChannelsFirst(self): 126 width = 7 127 channels = 3 128 images = random_ops.random_uniform((5, channels, width)) 129 layer = pooling_layers.AveragePooling1D( 130 2, strides=2, data_format='channels_first') 131 output = layer.apply(images) 132 self.assertListEqual(output.get_shape().as_list(), 133 [5, channels, width // 2]) 134 135 def testCreateMaxPooling3D(self): 136 depth, height, width = 6, 7, 9 137 images = random_ops.random_uniform((5, depth, height, width, 4)) 138 layer = pooling_layers.MaxPooling3D([2, 2, 2], strides=2) 139 output = layer.apply(images) 140 self.assertListEqual(output.get_shape().as_list(), [5, 3, 3, 4, 4]) 141 142 def testCreateAveragePooling3D(self): 143 depth, height, width = 6, 7, 9 144 images = random_ops.random_uniform((5, depth, height, width, 4)) 145 layer = pooling_layers.AveragePooling3D([2, 2, 2], strides=2) 146 output = layer.apply(images) 147 self.assertListEqual(output.get_shape().as_list(), [5, 3, 3, 4, 4]) 148 149 def testMaxPooling3DChannelsFirst(self): 150 depth, height, width = 6, 7, 9 151 images = random_ops.random_uniform((5, 2, depth, height, width)) 152 layer = pooling_layers.MaxPooling3D( 153 [2, 2, 2], strides=2, data_format='channels_first') 154 output = layer.apply(images) 155 self.assertListEqual(output.get_shape().as_list(), [5, 2, 3, 3, 4]) 156 157 def testAveragePooling3DChannelsFirst(self): 158 depth, height, width = 6, 7, 9 159 images = random_ops.random_uniform((5, 2, depth, height, width)) 160 layer = pooling_layers.AveragePooling3D( 161 [2, 2, 2], strides=2, data_format='channels_first') 162 output = layer.apply(images) 163 self.assertListEqual(output.get_shape().as_list(), [5, 2, 3, 3, 4]) 164 165 def testCreateMaxPooling2DIntegerPoolSize(self): 166 height, width = 7, 9 167 images = random_ops.random_uniform((5, height, width, 4)) 168 layer = pooling_layers.MaxPooling2D(2, strides=2) 169 output = layer.apply(images) 170 self.assertListEqual(output.get_shape().as_list(), [5, 3, 4, 4]) 171 172 def testMaxPooling2DPaddingSame(self): 173 height, width = 7, 9 174 images = random_ops.random_uniform((5, height, width, 4), seed=1) 175 layer = pooling_layers.MaxPooling2D( 176 images.get_shape()[1:3], strides=2, padding='same') 177 output = layer.apply(images) 178 self.assertListEqual(output.get_shape().as_list(), [5, 4, 5, 4]) 179 180 def testCreatePooling2DWithStrides(self): 181 height, width = 6, 8 182 # Test strides tuple 183 images = random_ops.random_uniform((5, height, width, 3), seed=1) 184 layer = pooling_layers.MaxPooling2D([2, 2], strides=(2, 2), padding='same') 185 output = layer.apply(images) 186 self.assertListEqual(output.get_shape().as_list(), 187 [5, height / 2, width / 2, 3]) 188 189 # Test strides integer 190 layer = pooling_layers.MaxPooling2D([2, 2], strides=2, padding='same') 191 output = layer.apply(images) 192 self.assertListEqual(output.get_shape().as_list(), 193 [5, height / 2, width / 2, 3]) 194 195 # Test unequal strides 196 layer = pooling_layers.MaxPooling2D([2, 2], strides=(2, 1), padding='same') 197 output = layer.apply(images) 198 self.assertListEqual(output.get_shape().as_list(), 199 [5, height / 2, width, 3]) 200 201 202 if __name__ == '__main__': 203 test.main() 204
