Update DPN

dual_path_network.py

@@ -254,10 +254,10 @@ def _bn_relu_conv_block(input, filters, kernel=(3, 3), stride=(1, 1), weight_dec
     '''
     channel_axis = 1 if K.image_data_format() == 'channels_first' else -1
 
-    x = BatchNormalization(axis=channel_axis)(input)
-    x = Activation('relu')(x)
     x = Conv2D(filters, kernel, padding='same', use_bias=False, kernel_initializer='he_normal',
-               kernel_regularizer=l2(weight_decay), strides=stride)(x)
+               kernel_regularizer=l2(weight_decay), strides=stride)(input)
+    x = BatchNormalization(axis=channel_axis)(x)
+    x = Activation('relu')(x)
     return x
 
 
@@ -278,15 +278,12 @@ def _grouped_convolution_block(input, grouped_channels, cardinality, strides, we
 
     if cardinality == 1:
         # with cardinality 1, it is a standard convolution
-        x = BatchNormalization(axis=channel_axis)(init)
-        x = Activation('relu')(x)
         x = Conv2D(grouped_channels, (3, 3), padding='same', use_bias=False, strides=strides,
-                   kernel_initializer='he_normal', kernel_regularizer=l2(weight_decay))(x)
+                   kernel_initializer='he_normal', kernel_regularizer=l2(weight_decay))(init)
+        x = BatchNormalization(axis=channel_axis)(x)
+        x = Activation('relu')(x)
         return x
 
-    input = BatchNormalization(axis=channel_axis)(init)
-    input = Activation('relu')(input)
-
     for c in range(cardinality):
         x = Lambda(lambda z: z[:, :, :, c * grouped_channels:(c + 1) * grouped_channels]
                    if K.image_data_format() == 'channels_last' else
@@ -298,6 +295,8 @@ def _grouped_convolution_block(input, grouped_channels, cardinality, strides, we
         group_list.append(x)
 
     group_merge = concatenate(group_list, axis=channel_axis)
+    group_merge = BatchNormalization(axis=channel_axis)(group_merge)
+    group_merge = Activation('relu')(group_merge)
     return group_merge
 
 
@@ -479,3 +478,7 @@ def _create_dpn(nb_classes, img_input, include_top, initial_conv_filters,
             x = Lambda(lambda z: 0.5 * z)(x)
 
     return x
+
+if __name__ == '__main__':
+    model = DPN92((224, 224, 3))
+    model.summary()