add gem layer

keras-team · Jul 3, 2022 · 54e15d5 · 54e15d5
1 parent 6bde237
commit 54e15d5
Show file tree

Hide file tree

Showing 8 changed files with 86 additions and 68 deletions.
diff --git a/keras/layers/__init__.py b/keras/layers/__init__.py
@@ -208,10 +208,16 @@
 from keras.layers.pooling.max_pooling3d import MaxPool3D
 from keras.layers.pooling.max_pooling3d import MaxPooling3D
 from keras.layers.pooling.generalized_mean_pooling1d import GeneralizedPooling1D
-from keras.layers.pooling.generalized_mean_pooling1d import GeneralizedMeanPooling1D
-from keras.layers.pooling.generalized_mean_pooling2d import GeneralizedMeanPooling2D
+from keras.layers.pooling.generalized_mean_pooling1d import (
+    GeneralizedMeanPooling1D,
+)
+from keras.layers.pooling.generalized_mean_pooling2d import (
+    GeneralizedMeanPooling2D,
+)
 from keras.layers.pooling.generalized_mean_pooling2d import GeneralizedPooling2D
-from keras.layers.pooling.generalized_mean_pooling3d import GeneralizedMeanPooling3D
+from keras.layers.pooling.generalized_mean_pooling3d import (
+    GeneralizedMeanPooling3D,
+)
 from keras.layers.pooling.generalized_mean_pooling3d import GeneralizedPooling3D
 from keras.layers.rnn.abstract_rnn_cell import AbstractRNNCell
 

diff --git a/keras/layers/pooling/__init__.py b/keras/layers/pooling/__init__.py
@@ -23,6 +23,18 @@
 from keras.layers.pooling.average_pooling2d import AvgPool2D
 from keras.layers.pooling.average_pooling3d import AveragePooling3D
 from keras.layers.pooling.average_pooling3d import AvgPool3D
+from keras.layers.pooling.generalized_mean_pooling1d import (
+    GeneralizedMeanPooling1D,
+)
+from keras.layers.pooling.generalized_mean_pooling1d import GeneralizedPooling1D
+from keras.layers.pooling.generalized_mean_pooling2d import (
+    GeneralizedMeanPooling2D,
+)
+from keras.layers.pooling.generalized_mean_pooling2d import GeneralizedPooling2D
+from keras.layers.pooling.generalized_mean_pooling3d import (
+    GeneralizedMeanPooling3D,
+)
+from keras.layers.pooling.generalized_mean_pooling3d import GeneralizedPooling3D
 from keras.layers.pooling.global_average_pooling1d import GlobalAveragePooling1D
 from keras.layers.pooling.global_average_pooling1d import GlobalAvgPool1D
 from keras.layers.pooling.global_average_pooling2d import GlobalAveragePooling2D
@@ -41,9 +53,3 @@
 from keras.layers.pooling.max_pooling2d import MaxPooling2D
 from keras.layers.pooling.max_pooling3d import MaxPool3D
 from keras.layers.pooling.max_pooling3d import MaxPooling3D
-from keras.layers.pooling.generalized_mean_pooling1d import GeneralizedPooling1D
-from keras.layers.pooling.generalized_mean_pooling1d import GeneralizedMeanPooling1D
-from keras.layers.pooling.generalized_mean_pooling2d import GeneralizedMeanPooling2D
-from keras.layers.pooling.generalized_mean_pooling2d import GeneralizedPooling2D
-from keras.layers.pooling.generalized_mean_pooling3d import GeneralizedMeanPooling3D
-from keras.layers.pooling.generalized_mean_pooling3d import GeneralizedPooling3D
diff --git a/keras/layers/pooling/base_generalized_pooling1d.py b/keras/layers/pooling/base_generalized_pooling1d.py
@@ -59,14 +59,14 @@ def __init__(
 
     def call(self, inputs):
         raise NotImplementedError
-    
+
     def get_config(self):
         config = {
             "power": self.power,
-            "pool_size": self.pool_size, 
+            "pool_size": self.pool_size,
             "strides": self.strides,
             "padding": self.padding,
-            "data_format": self.data_format
-            }
+            "data_format": self.data_format,
+        }
         base_config = super().get_config()
         return dict(list(base_config.items()) + list(config.items()))
diff --git a/keras/layers/pooling/base_generalized_pooling2d.py b/keras/layers/pooling/base_generalized_pooling2d.py
@@ -63,10 +63,10 @@ def call(self, inputs):
     def get_config(self):
         config = {
             "power": self.power,
-            "pool_size": self.pool_size, 
+            "pool_size": self.pool_size,
             "strides": self.strides,
             "padding": self.padding,
-            "data_format": self.data_format
-            }
+            "data_format": self.data_format,
+        }
         base_config = super().get_config()
         return dict(list(base_config.items()) + list(config.items()))
diff --git a/keras/layers/pooling/base_generalized_pooling3d.py b/keras/layers/pooling/base_generalized_pooling3d.py
@@ -59,14 +59,14 @@ def __init__(
 
     def call(self, inputs):
         raise NotImplementedError
-    
+
     def get_config(self):
         config = {
             "power": self.power,
-            "pool_size": self.pool_size, 
+            "pool_size": self.pool_size,
             "strides": self.strides,
             "padding": self.padding,
-            "data_format": self.data_format
-            }
+            "data_format": self.data_format,
+        }
         base_config = super().get_config()
         return dict(list(base_config.items()) + list(config.items()))
diff --git a/keras/layers/pooling/generalized_mean_pooling1d.py b/keras/layers/pooling/generalized_mean_pooling1d.py
@@ -9,13 +9,14 @@ class GeneralizedMeanPooling1D(GeneralizedPooling1D):
     """Generalized mean pooling operation for temporal data.
 
     Generalized Mean Pooling (GeM) computes the generalized mean of each
-    channel in a tensor. It provides a parameter `p` that sets an exponent
-    enabling the pooling to increase or decrease the contrast between salient
-    features in the feature map.
+    channel in a tensor. It provides a parameter `power` that sets an
+    exponent enabling the pooling to increase or decrease the contrast
+    between salient features in the feature map.
 
-    The GeM layer is an generalization of the average pooling layer and spatial
-    max pooling layer. When `p` = 1`, it will act as a average pooling layer and
-    when `p = inf`, it will act as a spatial max pooling layer.
+    The GeM layer is an generalization of the average pooling layer and
+    spatial max pooling layer. When `power` = 1`, it will act as a average
+    pooling layer and when `power = inf`, it will act as a spatial
+    max-pooling layer.
 
     Examples:
 
@@ -24,7 +25,8 @@ class GeneralizedMeanPooling1D(GeneralizedPooling1D):
     >>> input_shape = (2, 3, 4)
     >>> x = tf.random.normal(input_shape)
     >>> gem_pool_1d = tf.keras.layers.GeneralizedMeanPooling1D(power=3,
-    ...    pool_size=2, strides=1, padding='valid', data_format='channels_last')
+    ...    pool_size=2, strides=1, padding='valid',
+    ...    data_format='channels_last')
     >>> gem_pool_1d(x)
     <tf.Tensor: shape=(1, 4, 1), dtype=float32, numpy=
     array([[[1.6509637],
@@ -37,7 +39,8 @@ class GeneralizedMeanPooling1D(GeneralizedPooling1D):
     >>> input_shape = (2, 3, 4)
     >>> x = tf.random.normal(input_shape)
     >>> gem_pool_1d = tf.keras.layers.GeneralizedMeanPooling1D(power=3,
-    ...    pool_size=2, strides=2, padding='valid', data_format='channels_last')
+    ...    pool_size=2, strides=2, padding='valid',
+    ...    data_format='channels_last')
     >>> gem_pool_1d(x)
     <tf.Tensor: shape=(1, 2, 1), dtype=float32, numpy=
     array([[[1.6509637],
@@ -48,7 +51,8 @@ class GeneralizedMeanPooling1D(GeneralizedPooling1D):
     >>> input_shape = (2, 3, 4)
     >>> x = tf.random.normal(input_shape)
     >>> gem_pool_1d = tf.keras.layers.GeneralizedMeanPooling1D(power=3,
-    ...    pool_size=2, strides=1, padding='same', data_format='channels_last')
+    ...    pool_size=2, strides=1, padding='same',
+    ...    data_format='channels_last')
     >>> gem_pool_1d(x)
     <tf.Tensor: shape=(1, 5, 1), dtype=float32, numpy=
     array([[[1.6509637],
@@ -62,23 +66,22 @@ class GeneralizedMeanPooling1D(GeneralizedPooling1D):
         the generalized mean pooling computation. Setting this exponent as
         power > 1 increases the contrast of the pooled feature map and focuses
         on the salient features of the image. GeM is a generalization of the
-        average pooling commonly used in classification networks (power = 1)
-        and of spatial max-pooling layer (power = inf).
+        average pooling when `power` = 1 and of spatial max-pooling layer when
+        `power` = inf or a large number.
       pool_size: Integer, size of the average pooling windows.
       strides: Integer, or None. Factor by which to downscale.
         E.g. 2 will halve the input.
         If None, it will default to `pool_size`.
-      padding: One of `"valid"` or `"same"` (case-insensitive).
-        `"valid"` means no padding. `"same"` results in padding evenly to
+      padding: A string. The padding method, either 'valid' or 'same'.
+        `'valid'` means no padding. `'same'` results in padding evenly to
         the left/right or up/down of the input such that output has the same
         height/width dimension as the input.
-      data_format: A string,
-        one of `channels_last` (default) or `channels_first`.
-        The ordering of the dimensions in the inputs.
+      data_format: A string, one of `channels_last` (default) or
+        `channels_first`. The ordering of the dimensions in the inputs.
         `channels_last` corresponds to inputs with shape
-        `(batch, steps, features)` while `channels_first`
-        corresponds to inputs with shape
-        `(batch, features, steps)`.
+        `(batch, steps, features)` while `channels_first` corresponds
+        to inputs with shape `(batch, features, steps)`.
+      name: A string, the name of the layer.
 
     Input shape:
       - If `data_format='channels_last'`:

diff --git a/keras/layers/pooling/generalized_mean_pooling2d.py b/keras/layers/pooling/generalized_mean_pooling2d.py
@@ -9,13 +9,14 @@ class GeneralizedMeanPooling2D(GeneralizedPooling2D):
     """Generalized mean pooling operation for temporal data.
 
     Generalized Mean Pooling (GeM) computes the generalized mean of each
-    channel in a tensor. It provides a parameter `p` that sets an exponent
-    enabling the pooling to increase or decrease the contrast between salient
-    features in the feature map.
+    channel in a tensor. It provides a parameter `power` that sets an
+    exponent enabling the pooling to increase or decrease the contrast
+    between salient features in the feature map.
 
-    The GeM layer is an generalization of the average pooling layer and spatial
-    max pooling layer. When `p` = 1`, it will act as a average pooling layer and
-    when `p = inf`, it will act as a spatial max pooling layer.
+    The GeM layer is an generalization of the average pooling layer and
+    spatial max pooling layer. When `power` = 1`, it will act as a average
+    pooling layer and when `power = inf`, it will act as a spatial
+    max-pooling layer.
 
     Examples:
 
@@ -26,7 +27,8 @@ class GeneralizedMeanPooling2D(GeneralizedPooling2D):
     ...                  [7., 8., 9.]])
     >>> x = tf.reshape(x, [1, 3, 3, 1])
     >>> gem_pool_2d = tf.keras.layers.GeneralizedMeanPooling2D(power=3,
-    ...    pool_size=2, strides=1, padding='valid', data_format='channels_last')
+    ...    pool_size=2, strides=1, padding='valid',
+    ...    data_format='channels_last')
     >>> gem_pool_2d(x)
     <tf.Tensor: shape=(1, 2, 2, 1), dtype=float32, numpy=
     array([[[[3.6717105],
@@ -42,7 +44,8 @@ class GeneralizedMeanPooling2D(GeneralizedPooling2D):
     ...                  [9., 10., 11., 12.]])
     >>> x = tf.reshape(x, [1, 3, 4, 1])
     >>> gem_pool_2d = tf.keras.layers.GeneralizedMeanPooling2D(power=3,
-    ...    pool_size=2, strides=2, padding='valid', data_format='channels_last')
+    ...    pool_size=2, strides=2, padding='valid',
+    ...    data_format='channels_last')
     >>> gem_pool_2d(x)
     <tf.Tensor: shape=(1, 1, 2, 1), dtype=float32, numpy=
     array([[[[4.4395204],
@@ -55,7 +58,8 @@ class GeneralizedMeanPooling2D(GeneralizedPooling2D):
     ...                  [7., 8., 9.]])
     >>> x = tf.reshape(x, [1, 3, 3, 1])
     >>> gem_pool_2d = tf.keras.layers.GeneralizedMeanPooling1D(power=3,
-    ...    pool_size=2, strides=1, padding='same', data_format='channels_last')
+    ...    pool_size=2, strides=1, padding='same',
+    ...    data_format='channels_last')
     >>> gem_pool_2d(x)
     <tf.Tensor: shape=(1, 3, 3, 1), dtype=float32, numpy=
     array([[[[3.6717105],
@@ -75,8 +79,8 @@ class GeneralizedMeanPooling2D(GeneralizedPooling2D):
         the generalized mean pooling computation. Setting this exponent as
         power > 1 increases the contrast of the pooled feature map and focuses
         on the salient features of the image. GeM is a generalization of the
-        average pooling commonly used in classification networks (power = 1)
-        and of spatial max-pooling layer (power = inf).
+        average pooling when `power` = 1 and of spatial max-pooling layer when
+        `power` = inf or a large number.
       pool_size: An integer or tuple/list of 2 integers:
         (pool_height, pool_width) specifying the size of the pooling window.
         Can be a single integer to specify the same value for all spatial
@@ -85,12 +89,11 @@ class GeneralizedMeanPooling2D(GeneralizedPooling2D):
         of the pooling operation. Can be a single integer to specify the same
         value for all spatial dimensions.
       padding: A string. The padding method, either 'valid' or 'same'.
-        Case-insensitive.
       data_format: A string, one of `channels_last` (default) or
         `channels_first`. The ordering of the dimensions in the inputs.
         `channels_last` corresponds to inputs with shape
-        `(batch, height, width, channels)` while `channels_first` corresponds
-        to inputs with shape `(batch, channels, height, width)`.
+        `(batch, height, width, channels)` while `channels_first`
+        corresponds to inputs with shape `(batch, channels, height, width)`.
       name: A string, the name of the layer.
 
     Input shape:

diff --git a/keras/layers/pooling/generalized_mean_pooling3d.py b/keras/layers/pooling/generalized_mean_pooling3d.py
@@ -8,14 +8,15 @@
 class GeneralizedMeanPooling3D(GeneralizedPooling3D):
     """Generalized mean pooling operation for temporal data.
 
-    Generalized Mean Pooling (GeM) computes the generalized mean of each
-    channel in a tensor. It provides a parameter `p` that sets an exponent
-    enabling the pooling to increase or decrease the contrast between salient
-    features in the feature map.
+    Generalized Mean Pooling (GeM) computes the generalized mean of
+    each channel in a tensor. It provides a parameter `p` that sets
+    an exponent enabling the pooling to increase or decrease the contrast
+    between salient features in the feature map.
 
-    The GeM layer is an generalization of the average pooling layer and spatial
-    max pooling layer. When `p` = 1`, it will act as a average pooling layer and
-    when `p = inf`, it will act as a spatial max pooling layer.
+    The GeM layer is an generalization of the average pooling layer and
+    spatial max pooling layer. When `power` = 1`, it will act as a average
+    pooling layer and when `power = inf`, it will act as a spatial
+    max-pooling layer.
 
     Examples:
 
@@ -63,24 +64,23 @@ class GeneralizedMeanPooling3D(GeneralizedPooling3D):
     Args:
       power: Float power > 0 is an inverse exponent parameter, used during
         the generalized mean pooling computation. Setting this exponent as
-        power > 1 increases the contrast of the pooled feature map and
-        focuses on the salient features of the image. GeM is a generalization
-        of the average pooling commonly used in classification networks
-        (power = 1) and of spatial max-pooling layer (power = inf).
+        power > 1 increases the contrast of the pooled feature map and focuses
+        on the salient features of the image. GeM is a generalization of the
+        average pooling when `power` = 1 and of spatial max-pooling layer when
+        `power` = inf or a large number.
       pool_size: An integer or tuple/list of 3 integers:
-        (pool_depth, pool_height, pool_width) specifying the size of the
+        `(pool_depth, pool_height, pool_width)` specifying the size of the
         pooling window. Can be a single integer to specify the same value for
         all spatial dimensions.
       strides: An integer or tuple/list of 3 integers, specifying the strides
         of the pooling operation. Can be a single integer to specify the same
         value for all spatial dimensions.
       padding: A string. The padding method, either 'valid' or 'same'.
-        Case-insensitive.
       data_format: A string, one of `channels_last` (default) or
         `channels_first`. The ordering of the dimensions in the inputs.
         `channels_last` corresponds to inputs with shape
-        `(batch, depth, height, width, channels)` while
-        `channels_first` corresponds to inputs with shape
+        `(batch, depth, height, width, channels)` while `channels_first`
+        corresponds to inputs with shape
         `(batch, channels, depth, height, width)`.
       name: A string, the name of the layer.