densenet dilation for segmentation

examples/cifar10_densenet.py

@@ -1,8 +1,8 @@
-'''
+"""
 Trains a DenseNet-40-12 model on the CIFAR-10 Dataset.
 
 Gets a 94.84% accuracy score after 100 epochs.
-'''
+"""
 from __future__ import absolute_import
 from __future__ import print_function
 from __future__ import division
@@ -33,8 +33,8 @@
 dropout_rate = 0.0  # 0.0 for data augmentation
 
 # Create the model (without loading weights)
-model = DenseNet(depth, nb_dense_block, growth_rate, nb_filter, dropout_rate=dropout_rate,
-                 input_shape=img_dim, weights=None)
+model = DenseNet(img_dim, depth, nb_dense_block, growth_rate, nb_filter,
+                 dropout_rate=dropout_rate, weights=None)
 print('Model created')
 
 model.summary()

examples/cifar10_ror.py

@@ -1,8 +1,8 @@
-'''
+"""
 Trains a Residual-of-Residual Network (WRN-40-2) model on the CIFAR-10 Dataset.
 
 Gets a 94.53% accuracy score after 150 epochs.
-'''
+"""
 
 import keras.callbacks as callbacks
 import keras.utils.np_utils as kutils

examples/cifar10_wide_resnet.py

@@ -1,11 +1,11 @@
-'''
+"""
 Trains a WRN-28-8 model on the CIFAR-10 Dataset.
 
 Performance is slightly less than the paper, since
 they use WRN-28-10 model (95.83%).
 
 Gets a 95.54% accuracy score after 300 epochs.
-'''
+"""
 from __future__ import absolute_import
 from __future__ import print_function
 from __future__ import division

keras_contrib/applications/ror.py

@@ -211,7 +211,7 @@ def __conv_block(input, nb_filters=32, k=1, dropout=0.0):
 
 
 def __create_pre_residual_of_residual(nb_classes, img_input, include_top, depth=28, width=1, dropout=0.0):
-    '''
+    """
     Creates a Residual Network of Residual Network with specified parameters
 
     Example : To create a Pre-RoR model, use k = 1
@@ -230,7 +230,7 @@ def __create_pre_residual_of_residual(nb_classes, img_input, include_top, depth=
                  Note : Generally not used in RoR
 
     Returns: a Keras Model
-    '''
+    """
 
     N = (depth - 4) // 6
 

keras_contrib/applications/wide_resnet.py

@@ -251,7 +251,7 @@ def ___conv4_block(input, k=1, dropout=0.0):
 
 
 def __create_wide_residual_network(nb_classes, img_input, include_top, depth=28, width=8, dropout=0.0):
-    ''' Creates a Wide Residual Network with specified parameters
+    """ Creates a Wide Residual Network with specified parameters
 
     Args:
         nb_classes: Number of output classes
@@ -265,7 +265,7 @@ def __create_wide_residual_network(nb_classes, img_input, include_top, depth=28,
         dropout: Adds dropout if value is greater than 0.0
 
     Returns:a Keras Model
-    '''
+    """
 
     N = (depth - 4) // 6
 

keras_contrib/backend/tensorflow_backend.py

@@ -27,14 +27,14 @@ def _preprocess_deconv_output_shape(x, shape, data_format):
 
 def conv2d(x, kernel, strides=(1, 1), padding='valid', data_format='channels_first',
            image_shape=None, filter_shape=None):
-    '''2D convolution.
+    """2D convolution.
     # Arguments
         kernel: kernel tensor.
         strides: strides tuple.
         padding: string, "same" or "valid".
         data_format: "tf" or "th". Whether to use Theano or TensorFlow dimension ordering
         in inputs/kernels/ouputs.
-    '''
+    """
     if padding == 'same':
         padding = 'SAME'
     elif padding == 'valid':
@@ -74,7 +74,7 @@ def deconv3d(x, kernel, output_shape, strides=(1, 1, 1),
              padding='valid',
              data_format='default',
              image_shape=None, filter_shape=None):
-    '''3D deconvolution (i.e. transposed convolution).
+    """3D deconvolution (i.e. transposed convolution).
 
     # Arguments
         x: input tensor.
@@ -91,7 +91,7 @@ def deconv3d(x, kernel, output_shape, strides=(1, 1, 1),
 
     # Raises
         ValueError: if `data_format` is neither `tf` or `th`.
-    '''
+    """
     if data_format == 'default':
         data_format = image_data_format()
     if data_format not in {'channels_first', 'channels_last'}:
@@ -111,7 +111,7 @@ def deconv3d(x, kernel, output_shape, strides=(1, 1, 1),
 
 def extract_image_patches(x, ksizes, ssizes, padding='same',
                           data_format='tf'):
-    '''
+    """
     Extract the patches from an image
     # Parameters
 
@@ -125,7 +125,7 @@ def extract_image_patches(x, ksizes, ssizes, padding='same',
         The (k_w,k_h) patches extracted
         TF ==> (batch_size,w,h,k_w,k_h,c)
         TH ==> (batch_size,w,h,c,k_w,k_h)
-    '''
+    """
     kernel = [1, ksizes[0], ksizes[1], 1]
     strides = [1, ssizes[0], ssizes[1], 1]
     padding = _preprocess_padding(padding)
@@ -144,7 +144,7 @@ def extract_image_patches(x, ksizes, ssizes, padding='same',
 
 
 def depth_to_space(input, scale, data_format=None):
-    ''' Uses phase shift algorithm to convert channels/depth for spatial resolution '''
+    """ Uses phase shift algorithm to convert channels/depth for spatial resolution """
     if data_format is None:
         data_format = image_data_format()
     data_format = data_format.lower()
@@ -155,6 +155,6 @@ def depth_to_space(input, scale, data_format=None):
 
 
 def moments(x, axes, shift=None, keep_dims=False):
-    ''' Wrapper over tensorflow backend call '''
+    """ Wrapper over tensorflow backend call """
 
     return tf.nn.moments(x, axes, shift=shift, keep_dims=keep_dims)

keras_contrib/backend/theano_backend.py

@@ -25,9 +25,9 @@
 
 def conv2d(x, kernel, strides=(1, 1), padding='valid', data_format='channels_first',
            image_shape=None, filter_shape=None):
-    '''
+    """
     padding: string, "same" or "valid".
-    '''
+    """
     if data_format not in {'channels_first', 'channels_last'}:
         raise Exception('Unknown data_format ' + str(data_format))
 
@@ -88,7 +88,7 @@ def int_or_none(value):
 def deconv3d(x, kernel, output_shape, strides=(1, 1, 1),
              padding='valid',
              data_format=None, filter_shape=None):
-    '''3D deconvolution (transposed convolution).
+    """3D deconvolution (transposed convolution).
 
     # Arguments
         kernel: kernel tensor.
@@ -98,7 +98,7 @@ def deconv3d(x, kernel, output_shape, strides=(1, 1, 1),
         data_format: "channels_last" or "channels_first".
             Whether to use Theano or TensorFlow dimension ordering
         in inputs/kernels/ouputs.
-    '''
+    """
     flip_filters = False
     if data_format is None:
         data_format = image_data_format()
@@ -136,7 +136,7 @@ def deconv3d(x, kernel, output_shape, strides=(1, 1, 1),
 
 
 def extract_image_patches(X, ksizes, strides, padding='valid', data_format='channels_first'):
-    '''
+    """
     Extract the patches from an image
     Parameters
     ----------
@@ -150,7 +150,7 @@ def extract_image_patches(X, ksizes, strides, padding='valid', data_format='chan
     The (k_w,k_h) patches extracted
     TF ==> (batch_size,w,h,k_w,k_h,c)
     TH ==> (batch_size,w,h,c,k_w,k_h)
-    '''
+    """
     patch_size = ksizes[1]
     if padding == 'same':
         padding = 'ignore_borders'
@@ -174,7 +174,7 @@ def extract_image_patches(X, ksizes, strides, padding='valid', data_format='chan
 
 
 def depth_to_space(input, scale, data_format=None):
-    ''' Uses phase shift algorithm to convert channels/depth for spatial resolution '''
+    """ Uses phase shift algorithm to convert channels/depth for spatial resolution """
     if data_format is None:
         data_format = image_data_format()
     data_format = data_format.lower()
@@ -191,7 +191,7 @@ def depth_to_space(input, scale, data_format=None):
 
 
 def moments(x, axes, shift=None, keep_dims=False):
-    ''' Calculates and returns the mean and variance of the input '''
+    """ Calculates and returns the mean and variance of the input """
 
     mean_batch = KTH.mean(x, axis=axes, keepdims=keep_dims)
     var_batch = KTH.var(x, axis=axes, keepdims=keep_dims)

tests/keras_contrib/applications/densenet_test.py

@@ -0,0 +1,35 @@
+import pytest
+import numpy as np
+from numpy.testing import assert_allclose
+
+from keras import backend as K
+from keras.optimizers import Adam
+from keras_contrib.applications.densenet import DenseNet
+from keras_contrib.applications.densenet import DenseNetFCN
+
+
+def test_densenet():
+    '''Tests if DenseNet Models can be created correctly
+    '''
+    densenet = DenseNet()
+    optimizer = Adam(lr=1e-3)
+    densenet.compile(loss='categorical_crossentropy',
+                     optimizer=optimizer,
+                     metrics=['accuracy'])
+
+    fcn_densenet = DenseNetFCN((32, 32, 3))
+    fcn_densenet.compile(loss='categorical_crossentropy',
+                         optimizer=optimizer,
+                         metrics=['accuracy'])
+
+    atrous_densenet = DenseNet(depth=None, nb_dense_block=4, growth_rate=12,
+                               nb_filter=16, nb_layers_per_block=4,
+                               weights=None, dilation_rate=2)
+
+    atrous_densenet.compile(loss='categorical_crossentropy',
+                            optimizer=optimizer,
+                            metrics=['accuracy'])
+
+
+if __name__ == '__main__':
+    pytest.main([__file__])