Python keras.layers.AveragePooling2D() Examples

def model_base_test_CNN(input_shape):
    model = Sequential()

    model.add(Conv2D(32, (3, 3), input_shape=input_shape, padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(5, 5), strides=(3, 3), padding='same'))

    model.add(Conv2D(64, (3, 3), padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(5, 5), strides=(3, 3), padding='same'))

    model.add(Conv2D(64, (3, 3), padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(5, 5), strides=(3, 3), padding='same'))

    model.add(Conv2D(128, (3, 3), padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(3, 3), strides=(2, 2), padding='same'))

    model.add(Flatten())

    return model.input, model.output

# 64x3 model 

def d_block(inp, fil, p = True):

    skip = Conv2D(fil, 1, padding = 'same', kernel_initializer = 'he_normal')(inp)

    out = Conv2D(filters = fil, kernel_size = 3, padding = 'same', kernel_initializer = 'he_normal')(inp)
    out = LeakyReLU(0.2)(out)

    out = Conv2D(filters = fil, kernel_size = 3, padding = 'same', kernel_initializer = 'he_normal')(out)
    out = LeakyReLU(0.2)(out)

    out = Conv2D(fil, 1, padding = 'same', kernel_initializer = 'he_normal')(out)

    out = add([out, skip])
    out = LeakyReLU(0.2)(out)

    if p:
        out = AveragePooling2D()(out)

    return out 

def avg_pool2d(h_kernel_size, h_stride):

    def compile_fn(di, dh):
        layer = layers.AveragePooling2D(pool_size=dh['kernel_size'],
                                        strides=(dh['stride'], dh['stride']),
                                        padding='same')

        def fn(di):
            return {'out': layer(di['in'])}

        return fn

    return siso_keras_module('AvgPool', compile_fn, {
        'kernel_size': h_kernel_size,
        'stride': h_stride,
    }) 

def avg_pool2d(h_kernel_size, h_stride):

    def compile_fn(di, dh):
        layer = layers.AveragePooling2D(pool_size=dh['kernel_size'],
                                        strides=(dh['stride'], dh['stride']),
                                        padding='same')

        def fn(di):
            return {'out': layer(di['in'])}

        return fn

    return siso_keras_module('AvgPool2D', compile_fn, {
        'kernel_size': h_kernel_size,
        'stride': h_stride,
    }) 

def add_new_last_layer(base_model, nb_classes):
    """Add last layer to the convnet
    
    Args:
        base_model: keras model excluding top
        nb_classes: # of classes
        
    Returns:
        new keras model with last layer
    """
    x = base_model.output
    x = AveragePooling2D((8, 8), border_mode='valid', name='avg_pool')(x)
    x = Dropout(0.4)(x)
    x = Flatten()(x)
    predictions = Dense(2, activation='softmax')(x)
    model = Model(input=base_model.input, output=predictions)
    return model 

def transition_block(x, reduction, name):
    """A transition block.

    # Arguments
        x: input tensor.
        reduction: float, compression rate at transition layers.
        name: string, block label.

    # Returns
        output tensor for the block.
    """
    bn_axis = 3 if K.image_data_format() == 'channels_last' else 1
    x = BatchNormalization(axis=bn_axis, epsilon=1.001e-5,
                           name=name + '_bn')(x)
    x = Activation('relu', name=name + '_relu')(x)
    x = Conv2D(int(K.int_shape(x)[bn_axis] * reduction), 1, use_bias=False,
               name=name + '_conv')(x)
    x = AveragePooling2D(2, strides=2, name=name + '_pool')(x)
    return x 

def transition_layer(input_tensor, numFilters, compressionFactor=1.0):

    numOutPutFilters = int(numFilters*compressionFactor)

    if K.image_data_format() == 'channels_last':
        bn_axis = -1
    else:
        bn_axis = 1

    x = BatchNormalization(axis=bn_axis)(input_tensor)
    x = Activation('relu')(x)

    x = Conv2D(numOutPutFilters, (1, 1), strides=(1, 1), padding='same', kernel_initializer='he_normal')(x)

    # downsampling
    x = AveragePooling2D((2, 2), strides=(2, 2), padding='valid', data_format='channels_last', name='')(x)

    return x, numOutPutFilters 

def transition_SE_layer(input_tensor, numFilters, compressionFactor=1.0, se_ratio=16):

    numOutPutFilters = int(numFilters*compressionFactor)

    if K.image_data_format() == 'channels_last':
        bn_axis = -1
    else:
        bn_axis = 1

    x = BatchNormalization(axis=bn_axis)(input_tensor)
    x = Activation('relu')(x)

    x = Conv2D(numOutPutFilters, (1, 1), strides=(1, 1), padding='same', kernel_initializer='he_normal')(x)

    # SE Block
    x = squeeze_excitation_block(x, ratio=se_ratio)
    #x = BatchNormalization(axis=bn_axis)(x)

    # downsampling
    x = AveragePooling2D((2, 2), strides=(2, 2), padding='valid', data_format='channels_last', name='')(x)

    #x = squeeze_excitation_block(x, ratio=se_ratio)

    return x, numOutPutFilters 

def transition_layer(input_tensor, numFilters, compressionFactor=1.0):

    numOutPutFilters = int(numFilters*compressionFactor)

    if K.image_data_format() == 'channels_last':
        bn_axis = -1
    else:
        bn_axis = 1

    x = BatchNormalization(axis=bn_axis)(input_tensor)
    x = Activation('relu')(x)

    x = Conv2D(numOutPutFilters, (1, 1), strides=(1, 1), padding='same', kernel_initializer='he_normal')(x)

    # downsampling
    x = AveragePooling2D((2, 2), strides=(2, 2), padding='valid', data_format='channels_last', name='')(x)

    return x, numOutPutFilters 

def transition_SE_layer(input_tensor, numFilters, compressionFactor=1.0, se_ratio=16):

    numOutPutFilters = int(numFilters*compressionFactor)

    if K.image_data_format() == 'channels_last':
        bn_axis = -1
    else:
        bn_axis = 1

    x = BatchNormalization(axis=bn_axis)(input_tensor)
    x = Activation('relu')(x)

    x = Conv2D(numOutPutFilters, (1, 1), strides=(1, 1), padding='same', kernel_initializer='he_normal')(x)

    # SE Block
    x = squeeze_excitation_block(x, ratio=se_ratio)
    #x = BatchNormalization(axis=bn_axis)(x)

    # downsampling
    x = AveragePooling2D((2, 2), strides=(2, 2), padding='valid', data_format='channels_last', name='')(x)

    #x = squeeze_excitation_block(x, ratio=se_ratio)

    return x, numOutPutFilters 

def model_base_64x3_CNN(input_shape):
    model = Sequential()

    model.add(Conv2D(64, (3, 3), input_shape=input_shape, padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(5, 5), strides=(3, 3), padding='same'))

    model.add(Conv2D(64, (3, 3), padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(5, 5), strides=(3, 3), padding='same'))

    model.add(Conv2D(64, (3, 3), padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(5, 5), strides=(3, 3), padding='same'))

    model.add(Flatten())

    return model.input, model.output

# 4 CNN layer model 

def model_base_4_CNN(input_shape):
    model = Sequential()

    model.add(Conv2D(64, (5, 5), input_shape=input_shape, padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(5, 5), strides=(3, 3), padding='same'))

    model.add(Conv2D(64, (5, 5), padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(5, 5), strides=(3, 3), padding='same'))

    model.add(Conv2D(128, (5, 5), padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(3, 3), strides=(2, 2), padding='same'))

    model.add(Conv2D(256, (3, 3), padding='same'))
    model.add(Activation('relu'))
    model.add(AveragePooling2D(pool_size=(3, 3), strides=(2, 2), padding='same'))

    model.add(Flatten())

    return model.input, model.output

# 5 CNN layer with residual connections model 

def One_vs_One_Inception(self, nOutput=2, input=[224, 224, 3]):
        """
            Builds a simple One_vs_One_Inception network with 2 inception layers (useful for ECOC models).
        """
        if len(input) == 3:
            input_shape = tuple([input[2]] + input[0:2])
        else:
            input_shape = tuple(input)

        self.model = Graph()
        # Input
        self.model.add_input(name='input', input_shape=input_shape)
        # Inception Ea
        out_Ea = self.__addInception('inceptionEa', 'input', 4, 2, 8, 2, 2, 2)
        # Inception Eb
        out_Eb = self.__addInception('inceptionEb', out_Ea, 2, 2, 4, 2, 1, 1)
        # Average Pooling    pool_size=(7,7)
        self.model.add_node(AveragePooling2D(pool_size=input_shape[1:], strides=(1, 1)), name='ave_pool/ECOC',
                            input=out_Eb)
        # Softmax
        self.model.add_node(Flatten(), name='loss_OnevsOne/classifier_flatten', input='ave_pool/ECOC')
        self.model.add_node(Dropout(0.5), name='loss_OnevsOne/drop', input='loss_OnevsOne/classifier_flatten')
        self.model.add_node(Dense(nOutput, activation='softmax'), name='loss_OnevsOne', input='loss_OnevsOne/drop')
        # Output
        self.model.add_output(name='loss_OnevsOne/output', input='loss_OnevsOne') 

def add_One_vs_One_Inception(self, input, input_shape, id_branch, nOutput=2, activation='softmax'):
        """
            Builds a simple One_vs_One_Inception network with 2 inception layers on the top of the current model (useful for ECOC_loss models).
        """

        # Inception Ea
        out_Ea = self.__addInception('inceptionEa_' + str(id_branch), input, 4, 2, 8, 2, 2, 2)
        # Inception Eb
        out_Eb = self.__addInception('inceptionEb_' + str(id_branch), out_Ea, 2, 2, 4, 2, 1, 1)
        # Average Pooling    pool_size=(7,7)
        self.model.add_node(AveragePooling2D(pool_size=input_shape[1:], strides=(1, 1)),
                            name='ave_pool/ECOC_' + str(id_branch), input=out_Eb)
        # Softmax
        self.model.add_node(Flatten(),
                            name='fc_OnevsOne_' + str(id_branch) + '/flatten', input='ave_pool/ECOC_' + str(id_branch))
        self.model.add_node(Dropout(0.5),
                            name='fc_OnevsOne_' + str(id_branch) + '/drop',
                            input='fc_OnevsOne_' + str(id_branch) + '/flatten')
        output_name = 'fc_OnevsOne_' + str(id_branch)
        self.model.add_node(Dense(nOutput, activation=activation),
                            name=output_name, input='fc_OnevsOne_' + str(id_branch) + '/drop')

        return output_name 

def add_One_vs_One_Inception_Functional(self, input, input_shape, id_branch, nOutput=2, activation='softmax'):
        """
            Builds a simple One_vs_One_Inception network with 2 inception layers on the top of the current model (useful for ECOC_loss models).
        """

        in_node = self.model.get_layer(input).output

        # Inception Ea
        [out_Ea, out_Ea_name] = self.__addInception_Functional('inceptionEa_' + str(id_branch), in_node, 4, 2, 8, 2, 2,
                                                               2)
        # Inception Eb
        [out_Eb, out_Eb_name] = self.__addInception_Functional('inceptionEb_' + str(id_branch), out_Ea, 2, 2, 4, 2, 1,
                                                               1)
        # Average Pooling    pool_size=(7,7)
        x = AveragePooling2D(pool_size=input_shape, strides=(1, 1), name='ave_pool/ECOC_' + str(id_branch))(out_Eb)

        # Softmax
        output_name = 'fc_OnevsOne_' + str(id_branch)
        x = Flatten(name='fc_OnevsOne_' + str(id_branch) + '/flatten')(x)
        x = Dropout(0.5, name='fc_OnevsOne_' + str(id_branch) + '/drop')(x)
        out_node = Dense(nOutput, activation=activation, name=output_name)(x)

        return out_node 

def One_vs_One_Inception_v2(self, nOutput=2, input=[224, 224, 3]):
        """
            Builds a simple One_vs_One_Inception_v2 network with 2 inception layers (useful for ECOC models).
        """
        if len(input) == 3:
            input_shape = tuple([input[2]] + input[0:2])
        else:
            input_shape = tuple(input)

        self.model = Graph()
        # Input
        self.model.add_input(name='input', input_shape=input_shape)
        # Inception Ea
        out_Ea = self.__addInception('inceptionEa', 'input', 16, 8, 32, 8, 8, 8)
        # Inception Eb
        out_Eb = self.__addInception('inceptionEb', out_Ea, 8, 8, 16, 8, 4, 4)
        # Average Pooling    pool_size=(7,7)
        self.model.add_node(AveragePooling2D(pool_size=input_shape[1:], strides=(1, 1)), name='ave_pool/ECOC',
                            input=out_Eb)
        # Softmax
        self.model.add_node(Flatten(), name='loss_OnevsOne/classifier_flatten', input='ave_pool/ECOC')
        self.model.add_node(Dropout(0.5), name='loss_OnevsOne/drop', input='loss_OnevsOne/classifier_flatten')
        self.model.add_node(Dense(nOutput, activation='softmax'), name='loss_OnevsOne', input='loss_OnevsOne/drop')
        # Output
        self.model.add_output(name='loss_OnevsOne/output', input='loss_OnevsOne') 

def One_vs_One_Inception(self, nOutput=2, input=[224, 224, 3]):
        """
            Builds a simple One_vs_One_Inception network with 2 inception layers (useful for ECOC models).
        """
        if len(input) == 3:
            input_shape = tuple([input[2]] + input[0:2])
        else:
            input_shape = tuple(input)

        self.model = Graph()
        # Input
        self.model.add_input(name='input', input_shape=input_shape)
        # Inception Ea
        out_Ea = self.__addInception('inceptionEa', 'input', 4, 2, 8, 2, 2, 2)
        # Inception Eb
        out_Eb = self.__addInception('inceptionEb', out_Ea, 2, 2, 4, 2, 1, 1)
        # Average Pooling    pool_size=(7,7)
        self.model.add_node(AveragePooling2D(pool_size=input_shape[1:], strides=(1, 1)), name='ave_pool/ECOC',
                            input=out_Eb)
        # Softmax
        self.model.add_node(Flatten(), name='loss_OnevsOne/classifier_flatten', input='ave_pool/ECOC')
        self.model.add_node(Dropout(0.5), name='loss_OnevsOne/drop', input='loss_OnevsOne/classifier_flatten')
        self.model.add_node(Dense(nOutput, activation='softmax'), name='loss_OnevsOne', input='loss_OnevsOne/drop')
        # Output
        self.model.add_output(name='loss_OnevsOne/output', input='loss_OnevsOne') 

def add_One_vs_One_Inception(self, input, input_shape, id_branch, nOutput=2, activation='softmax'):
        """
            Builds a simple One_vs_One_Inception network with 2 inception layers on the top of the current model (useful for ECOC_loss models).
        """

        # Inception Ea
        out_Ea = self.__addInception('inceptionEa_' + str(id_branch), input, 4, 2, 8, 2, 2, 2)
        # Inception Eb
        out_Eb = self.__addInception('inceptionEb_' + str(id_branch), out_Ea, 2, 2, 4, 2, 1, 1)
        # Average Pooling    pool_size=(7,7)
        self.model.add_node(AveragePooling2D(pool_size=input_shape[1:], strides=(1, 1)),
                            name='ave_pool/ECOC_' + str(id_branch), input=out_Eb)
        # Softmax
        self.model.add_node(Flatten(),
                            name='fc_OnevsOne_' + str(id_branch) + '/flatten', input='ave_pool/ECOC_' + str(id_branch))
        self.model.add_node(Dropout(0.5),
                            name='fc_OnevsOne_' + str(id_branch) + '/drop',
                            input='fc_OnevsOne_' + str(id_branch) + '/flatten')
        output_name = 'fc_OnevsOne_' + str(id_branch)
        self.model.add_node(Dense(nOutput, activation=activation),
                            name=output_name, input='fc_OnevsOne_' + str(id_branch) + '/drop')

        return output_name 

def add_One_vs_One_Inception_Functional(self, input, input_shape, id_branch, nOutput=2, activation='softmax'):
        """
            Builds a simple One_vs_One_Inception network with 2 inception layers on the top of the current model (useful for ECOC_loss models).
        """

        in_node = self.model.get_layer(input).output

        # Inception Ea
        [out_Ea, out_Ea_name] = self.__addInception_Functional('inceptionEa_' + str(id_branch), in_node, 4, 2, 8, 2, 2,
                                                               2)
        # Inception Eb
        [out_Eb, out_Eb_name] = self.__addInception_Functional('inceptionEb_' + str(id_branch), out_Ea, 2, 2, 4, 2, 1,
                                                               1)
        # Average Pooling    pool_size=(7,7)
        x = AveragePooling2D(pool_size=input_shape, strides=(1, 1), name='ave_pool/ECOC_' + str(id_branch))(out_Eb)

        # Softmax
        output_name = 'fc_OnevsOne_' + str(id_branch)
        x = Flatten(name='fc_OnevsOne_' + str(id_branch) + '/flatten')(x)
        x = Dropout(0.5, name='fc_OnevsOne_' + str(id_branch) + '/drop')(x)
        out_node = Dense(nOutput, activation=activation, name=output_name)(x)

        return out_node 

def add_One_vs_One_Inception_v2(self, input, input_shape, id_branch, nOutput=2, activation='softmax'):
        """
            Builds a simple One_vs_One_Inception_v2 network with 2 inception layers on the top of the current model (useful for ECOC_loss models).
        """

        # Inception Ea
        out_Ea = self.__addInception('inceptionEa_' + str(id_branch), input, 16, 8, 32, 8, 8, 8)
        # Inception Eb
        out_Eb = self.__addInception('inceptionEb_' + str(id_branch), out_Ea, 8, 8, 16, 8, 4, 4)
        # Average Pooling    pool_size=(7,7)
        self.model.add_node(AveragePooling2D(pool_size=input_shape[1:], strides=(1, 1)),
                            name='ave_pool/ECOC_' + str(id_branch), input=out_Eb)
        # Softmax
        self.model.add_node(Flatten(),
                            name='fc_OnevsOne_' + str(id_branch) + '/flatten', input='ave_pool/ECOC_' + str(id_branch))
        self.model.add_node(Dropout(0.5),
                            name='fc_OnevsOne_' + str(id_branch) + '/drop',
                            input='fc_OnevsOne_' + str(id_branch) + '/flatten')
        output_name = 'fc_OnevsOne_' + str(id_branch)
        self.model.add_node(Dense(nOutput, activation=activation),
                            name=output_name, input='fc_OnevsOne_' + str(id_branch) + '/drop')

        return output_name 

def apn_module(self, x):

        def right(x):
            x = layers.AveragePooling2D()(x)
            x = layers.Conv2D(self.classes, kernel_size=1, padding='same')(x)
            x = layers.BatchNormalization()(x)
            x = layers.Activation('relu')(x)
            x = layers.UpSampling2D(interpolation='bilinear')(x)
            return x

        def conv(x, filters, kernel_size, stride):
            x = layers.Conv2D(filters, kernel_size=kernel_size, strides=(stride, stride), padding='same')(x)
            x = layers.BatchNormalization()(x)
            x = layers.Activation('relu')(x)
            return x

        x_7 = conv(x, int(x.shape[-1]), 7, stride=2)
        x_5 = conv(x_7, int(x.shape[-1]), 5, stride=2)
        x_3 = conv(x_5, int(x.shape[-1]), 3, stride=2)

        x_3_1 = conv(x_3, self.classes, 3, stride=1)
        x_3_1_up = layers.UpSampling2D(interpolation='bilinear')(x_3_1)
        x_5_1 = conv(x_5, self.classes, 5, stride=1)
        x_3_5 = layers.add([x_5_1, x_3_1_up])
        x_3_5_up = layers.UpSampling2D(interpolation='bilinear')(x_3_5)
        x_7_1 = conv(x_7, self.classes, 3, stride=1)
        x_3_5_7 = layers.add([x_7_1, x_3_5_up])
        x_3_5_7_up = layers.UpSampling2D(interpolation='bilinear')(x_3_5_7)

        x_middle = conv(x, self.classes, 1, stride=1)
        x_middle = layers.multiply([x_3_5_7_up, x_middle])

        x_right = right(x)
        x_middle = layers.add([x_middle, x_right])
        return x_middle 

def psp_block(prev_layer, level, feature_map_shape, input_shape):
    if input_shape == (512, 512):
        kernel_strides_map = {1: [64, 64],
                              2: [32, 32],
                              3: [22, 21],
                              6: [11, 9]}  # TODO: Level 6: Kernel correct, but stride not exactly the same as Pytorch
    else:
        raise ValueError("Pooling parameters for input shape " + input_shape + " are not defined.")

    if K.image_data_format() == 'channels_last':
        bn_axis = 3
    else:
        bn_axis = 1

    names = [
        "class_psp_" + str(level) + "_conv",
        "class_psp_" + str(level) + "_bn"
    ]
    kernel = (kernel_strides_map[level][0], kernel_strides_map[level][0])
    strides = (kernel_strides_map[level][1], kernel_strides_map[level][1])
    prev_layer = AveragePooling2D(kernel, strides=strides)(prev_layer)
    prev_layer = Conv2D(512, (1, 1), strides=(1, 1), name=names[0], use_bias=False)(prev_layer)
    prev_layer = resnet.BN(bn_axis, name=names[1])(prev_layer)
    prev_layer = Activation('relu')(prev_layer)
    prev_layer = Upsampling(feature_map_shape)(prev_layer)
    return prev_layer 

def interp_block(x, num_filters=512, level=1, input_shape=(512, 512, 3), output_stride=16):
    feature_map_shape = (input_shape[0] / output_stride, input_shape[1] / output_stride)

    # compute dataformat
    if K.image_data_format() == 'channels_last':
        bn_axis = 3
    else:
        bn_axis = 1

    if output_stride == 16:
        scale = 5
    elif output_stride == 8:
        scale = 10

    kernel = (level*scale, level*scale)
    strides = (level*scale, level*scale)
    global_feat = AveragePooling2D(kernel, strides=strides, name='pool_level_%s_%s'%(level, output_stride))(x)
    global_feat = _conv(
            filters=num_filters,
            kernel_size=(1, 1),
            padding='same',
            name='conv_level_%s_%s'%(level,output_stride))(global_feat)
    global_feat = BatchNormalization(axis=bn_axis, name='bn_level_%s_%s'%(level, output_stride))(global_feat)
    global_feat = Lambda(Interp, arguments={'shape': feature_map_shape})(global_feat)

    return global_feat


# squeeze and excitation function 

def interp_block(prev_layer, level, feature_map_shape, input_shape):
    if input_shape == (473, 473):
        kernel_strides_map = {1: 60,
                              2: 30,
                              3: 20,
                              6: 10}
    elif input_shape == (713, 713):
        kernel_strides_map = {1: 90,
                              2: 45,
                              3: 30,
                              6: 15}
    else:
        print("Pooling parameters for input shape ",
              input_shape, " are not defined.")
        exit(1)

    names = [
        "conv5_3_pool" + str(level) + "_conv",
        "conv5_3_pool" + str(level) + "_conv_bn"
    ]
    kernel = (kernel_strides_map[level], kernel_strides_map[level])
    strides = (kernel_strides_map[level], kernel_strides_map[level])
    prev_layer = AveragePooling2D(kernel, strides=strides)(prev_layer)
    prev_layer = Conv2D(512, (1, 1), strides=(1, 1), name=names[0],
                        use_bias=False)(prev_layer)
    prev_layer = BN(name=names[1])(prev_layer)
    prev_layer = Activation('relu')(prev_layer)
    # prev_layer = Lambda(Interp, arguments={
    #                    'shape': feature_map_shape})(prev_layer)
    prev_layer = Interp(feature_map_shape)(prev_layer)
    return prev_layer 

def interp_block(prev_layer, level, feature_map_shape, input_shape):
    if input_shape == (473, 473):
        kernel_strides_map = {1: 60,
                              2: 30,
                              3: 20,
                              6: 10}
    elif input_shape == (713, 713):
        kernel_strides_map = {1: 90,
                              2: 45,
                              3: 30,
                              6: 15}
    else:
        print("Pooling parameters for input shape ",
              input_shape, " are not defined.")
        exit(1)

    names = [
        "conv5_3_pool" + str(level) + "_conv",
        "conv5_3_pool" + str(level) + "_conv_bn"
    ]
    kernel = (kernel_strides_map[level], kernel_strides_map[level])
    strides = (kernel_strides_map[level], kernel_strides_map[level])
    prev_layer = AveragePooling2D(kernel, strides=strides)(prev_layer)
    prev_layer = Conv2D(512, (1, 1), strides=(1, 1), name=names[0],
                        use_bias=False)(prev_layer)
    prev_layer = BN(name=names[1])(prev_layer)
    prev_layer = Activation('relu')(prev_layer)
    # prev_layer = Lambda(Interp, arguments={
    #                    'shape': feature_map_shape})(prev_layer)
    prev_layer = Interp(feature_map_shape)(prev_layer)
    return prev_layer 

def vgg_norm():
    img_input = Input(shape=(256, 256, 3))
    x1 = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv1')(img_input)
    x2 = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv2')(x1)
    x3 = AveragePooling2D((2, 2), strides=(2, 2), name='block1_pool')(x2)

    x4 = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv1')(x3)
    x5 = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv2')(x4)
    x6 = AveragePooling2D((2, 2), strides=(2, 2), name='block2_pool')(x5)

    x7 = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv1')(x6)
    x8 = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv2')(x7)
    x9 = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv3')(x8)
    x10 = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv4')(x9)
    x11 = AveragePooling2D((2, 2), strides=(2, 2), name='block3_pool')(x10)

    x12 = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv1')(x11)
    x13 = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv2')(x12)
    x14 = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv3')(x13)
    x15 = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv4')(x14)
    x16 = AveragePooling2D((2, 2), strides=(2, 2), name='block4_pool')(x15)

    x17 = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv1')(x16)
    x18 = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv2')(x17)
    x19 = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv3')(x18)
    x20 = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv4')(x19)
    x21 = AveragePooling2D((2, 2), strides=(2, 2), name='block5_pool')(x20)

    model = Model(inputs=[img_input], outputs=[x1, x2, x4, x5, x7, x8, x9, x10, x12, x13, x14, x15])
    model_orig = VGG19(weights='imagenet', input_shape=(256, 256, 3), include_top=False)

    for i in range(len(model.layers)):
        weights = model_orig.layers[i].get_weights()
        model.layers[i].set_weights(weights)

    return model 

def create_model():
    #Data format:tensorflow,channels_last;theano,channels_last
    if DATA_FORMAT=='channels_first':
        INP_SHAPE=(3,299,299)
        img_input=Input(shape=INP_SHAPE)
        CONCAT_AXIS=1
    elif DATA_FORMAT=='channels_last':
        INP_SHAPE=(299,299,3)
        img_input=Input(shape=INP_SHAPE)
        CONCAT_AXIS=3
    else:
        raise Exception('Invalid Dim Ordering')
    base_model = InceptionV3(weights='imagenet', include_top=False)
    base_model.summary()
    for layer in base_model.layers:
        layer.trainable = False

    x =  base_model.get_layer('mixed7').output
       

    x = Convolution2D(512, (1, 1), kernel_initializer="glorot_uniform", padding="same", name="DenseNet_initial_conv2D", use_bias=False,
                      kernel_regularizer=l2(WEIGHT_DECAY))(x)

    x = BatchNormalization()(x)

    x, nb_filter = dense_block(x, 5, 512, growth_rate=64,dropout_rate=0.5)

    x = AveragePooling2D(pool_size=(7, 7), strides=1, padding='valid', data_format=DATA_FORMAT)(x)

    x = Dense(512, activation='relu')(x)
    #x = Dropout(0.5)(x)
    x = Dense(16)(x)

    x = Lambda(lambda x:tf.nn.l2_normalize(x))(x)

    model = Model(inputs=base_model.input, outputs=x)

    return model 

def add_transition(input, num_filter = 12, dropout_rate = 0.2):
    global weight_decay
    BatchNorm = BatchNormalization()(input)
    relu = Activation('relu')(BatchNorm)
    Conv2D_BottleNeck = Conv2D(int(num_filter*compression), (1,1), use_bias=False, padding='same', kernel_initializer='he_normal', kernel_regularizer=l2(weight_decay))(relu)
    if dropout_rate>0:
      Conv2D_BottleNeck = Dropout(dropout_rate)(Conv2D_BottleNeck)
    avg = AveragePooling2D(pool_size=(2,2))(Conv2D_BottleNeck)
    
    return avg, int(num_filter*compression) 

def output_layer(input):
    global compression
    BatchNorm = BatchNormalization()(input)
    relu = Activation('relu')(BatchNorm)
    AvgPooling = AveragePooling2D(pool_size=(2,2))(relu)
    flat = Flatten()(AvgPooling)
    output = Dense(num_classes, activation='softmax')(flat)
    
    return output 

def modelMultiScaleDiscriminator(self, name=None):
        x1 = Input(shape=self.img_shape)
        x2 = AveragePooling2D(pool_size=(2, 2))(x1)
        #x4 = AveragePooling2D(pool_size=(2, 2))(x2)

        out_x1 = self.modelDiscriminator('D1')(x1)
        out_x2 = self.modelDiscriminator('D2')(x2)
        #out_x4 = self.modelDiscriminator('D4')(x4)

        return Model(inputs=x1, outputs=[out_x1, out_x2], name=name)