Python keras.layers.Permute() Examples

def softmax(x, restore_shape=True):
    """
    Softmax activation for a tensor x. No need to unroll the input first.
    :param x: x is a tensor with shape (None, channels, h, w)
    :param restore_shape: if False, output is returned unrolled (None, h * w, channels)
    :return: softmax activation of tensor x
    """
    _, c, h, w = x._keras_shape
    x = Permute(dims=(2, 3, 1))(x)
    x = Reshape(target_shape=(h * w, c))(x)

    x = Activation('softmax')(x)

    if restore_shape:
        x = Reshape(target_shape=(h, w, c))(x)
        x = Permute(dims=(3, 1, 2))(x)

    return x 

def model_definition():
        """ Keras ONetwork for MTCNN """
        input_ = Input(shape=(48, 48, 3))
        var_x = Conv2D(32, (3, 3), strides=1, padding='valid', name='conv1')(input_)
        var_x = PReLU(shared_axes=[1, 2], name='prelu1')(var_x)
        var_x = MaxPool2D(pool_size=3, strides=2, padding='same')(var_x)
        var_x = Conv2D(64, (3, 3), strides=1, padding='valid', name='conv2')(var_x)
        var_x = PReLU(shared_axes=[1, 2], name='prelu2')(var_x)
        var_x = MaxPool2D(pool_size=3, strides=2)(var_x)
        var_x = Conv2D(64, (3, 3), strides=1, padding='valid', name='conv3')(var_x)
        var_x = PReLU(shared_axes=[1, 2], name='prelu3')(var_x)
        var_x = MaxPool2D(pool_size=2)(var_x)
        var_x = Conv2D(128, (2, 2), strides=1, padding='valid', name='conv4')(var_x)
        var_x = PReLU(shared_axes=[1, 2], name='prelu4')(var_x)
        var_x = Permute((3, 2, 1))(var_x)
        var_x = Flatten()(var_x)
        var_x = Dense(256, name='conv5')(var_x)
        var_x = PReLU(name='prelu5')(var_x)

        classifier = Dense(2, activation='softmax', name='conv6-1')(var_x)
        bbox_regress = Dense(4, name='conv6-2')(var_x)
        landmark_regress = Dense(10, name='conv6-3')(var_x)
        return [input_], [classifier, bbox_regress, landmark_regress] 

def interp_net():
    if gpu_num > 1:
        dev = "/cpu:0"
    else:
        dev = "/gpu:0"
    with tf.device(dev):
        main_input = Input(shape=(4*num_features, timestamp), name='input')
        sci = single_channel_interp(ref_points, hours_look_ahead)
        cci = cross_channel_interp()
        interp = cci(sci(main_input))
        reconst = cci(sci(main_input, reconstruction=True),
                      reconstruction=True)
        aux_output = Lambda(lambda x: x, name='aux_output')(reconst)
        z = Permute((2, 1))(interp)
        z = GRU(hid, activation='tanh', recurrent_dropout=0.2, dropout=0.2)(z)
        main_output = Dense(1, activation='sigmoid', name='main_output')(z)
        orig_model = Model([main_input], [main_output, aux_output])
    if gpu_num > 1:
        model = multi_gpu_model(orig_model, gpus=gpu_num)
    else:
        model = orig_model
    print(orig_model.summary())
    return model 

def softmax_by_row(self, z: typing.Any) -> tuple:
        """Conduct softmax on each dimension across the four gates."""

        # z_transform: [B, U, 4]
        z_transform = Permute((2, 1))(Reshape((4, self._units))(z))
        size = [-1, 1, -1]
        # Perform softmax on each slice
        for i in range(0, self._units):
            begin = [0, i, 0]
            # z_slice: [B, 1, 4]
            z_slice = tf.slice(z_transform, begin, size)
            if i == 0:
                z_s = tf.nn.softmax(z_slice)
            else:
                z_s = tf.concat([z_s, tf.nn.softmax(z_slice)], 1)
        # zi, zl, zt, zd: [B, U]
        zi, zl, zt, zd = tf.unstack(z_s, axis=2)
        return zi, zl, zt, zd 

def create_Kao_Onet( weight_path = 'model48.h5'):
    input = Input(shape = [48,48,3])
    x = Conv2D(32, (3, 3), strides=1, padding='valid', name='conv1')(input)
    x = PReLU(shared_axes=[1,2],name='prelu1')(x)
    x = MaxPool2D(pool_size=3, strides=2, padding='same')(x)
    x = Conv2D(64, (3, 3), strides=1, padding='valid', name='conv2')(x)
    x = PReLU(shared_axes=[1,2],name='prelu2')(x)
    x = MaxPool2D(pool_size=3, strides=2)(x)
    x = Conv2D(64, (3, 3), strides=1, padding='valid', name='conv3')(x)
    x = PReLU(shared_axes=[1,2],name='prelu3')(x)
    x = MaxPool2D(pool_size=2)(x)
    x = Conv2D(128, (2, 2), strides=1, padding='valid', name='conv4')(x)
    x = PReLU(shared_axes=[1,2],name='prelu4')(x)
    x = Permute((3,2,1))(x)
    x = Flatten()(x)
    x = Dense(256, name='conv5') (x)
    x = PReLU(name='prelu5')(x)

    classifier = Dense(2, activation='softmax',name='conv6-1')(x)
    bbox_regress = Dense(4,name='conv6-2')(x)
    landmark_regress = Dense(10,name='conv6-3')(x)
    model = Model([input], [classifier, bbox_regress, landmark_regress])
    model.load_weights(weight_path, by_name=True)

    return model 

def create_Kao_Rnet (weight_path = 'model24.h5'):
    input = Input(shape=[24, 24, 3])  # change this shape to [None,None,3] to enable arbitraty shape input
    x = Conv2D(28, (3, 3), strides=1, padding='valid', name='conv1')(input)
    x = PReLU(shared_axes=[1, 2], name='prelu1')(x)
    x = MaxPool2D(pool_size=3,strides=2, padding='same')(x)

    x = Conv2D(48, (3, 3), strides=1, padding='valid', name='conv2')(x)
    x = PReLU(shared_axes=[1, 2], name='prelu2')(x)
    x = MaxPool2D(pool_size=3, strides=2)(x)

    x = Conv2D(64, (2, 2), strides=1, padding='valid', name='conv3')(x)
    x = PReLU(shared_axes=[1, 2], name='prelu3')(x)
    x = Permute((3, 2, 1))(x)
    x = Flatten()(x)
    x = Dense(128, name='conv4')(x)
    x = PReLU( name='prelu4')(x)
    classifier = Dense(2, activation='softmax', name='conv5-1')(x)
    bbox_regress = Dense(4, name='conv5-2')(x)
    model = Model([input], [classifier, bbox_regress])
    model.load_weights(weight_path, by_name=True)
    return model 

def model_definition():
        """ Keras RNetwork for MTCNN """
        input_ = Input(shape=(24, 24, 3))
        var_x = Conv2D(28, (3, 3), strides=1, padding='valid', name='conv1')(input_)
        var_x = PReLU(shared_axes=[1, 2], name='prelu1')(var_x)
        var_x = MaxPool2D(pool_size=3, strides=2, padding='same')(var_x)

        var_x = Conv2D(48, (3, 3), strides=1, padding='valid', name='conv2')(var_x)
        var_x = PReLU(shared_axes=[1, 2], name='prelu2')(var_x)
        var_x = MaxPool2D(pool_size=3, strides=2)(var_x)

        var_x = Conv2D(64, (2, 2), strides=1, padding='valid', name='conv3')(var_x)
        var_x = PReLU(shared_axes=[1, 2], name='prelu3')(var_x)
        var_x = Permute((3, 2, 1))(var_x)
        var_x = Flatten()(var_x)
        var_x = Dense(128, name='conv4')(var_x)
        var_x = PReLU(name='prelu4')(var_x)
        classifier = Dense(2, activation='softmax', name='conv5-1')(var_x)
        bbox_regress = Dense(4, name='conv5-2')(var_x)
        return [input_], [classifier, bbox_regress] 

def softmax(x, restore_shape=True):
    """
    Softmax activation for a tensor x. No need to unroll the input first.

    :param x: x is a tensor with shape (None, channels, h, w)
    :param restore_shape: if False, output is returned unrolled (None, h * w, channels)
    :return: softmax activation of tensor x
    """
    _, c, h, w = x._keras_shape
    x = Permute(dims=(2, 3, 1))(x)
    x = Reshape(target_shape=(h * w, c))(x)

    x = Activation('softmax')(x)

    if restore_shape:
        x = Reshape(target_shape=(h, w, c))(x)
        x = Permute(dims=(3, 1, 2))(x)

    return x 

def DUC(factor=(8, 8)):

    if factor[0] != factor[1]:
        raise ValueError('DUC upconvolution support only equal factors, '
                         'got {}'.format(factor))
    factor = factor[0]

    def layer(input_tensor):

        h, w, c = int_shape(input_tensor)[1:]
        H = h * factor
        W = w * factor

        x = Conv2DBlock(c*factor**2, (1,1),
                        padding='same',
                        name='duc_{}'.format(factor))(input_tensor)
        x = Permute((3, 1, 2))(x)
        x = Reshape((c, factor, factor, h, w))(x)
        x = Permute((1, 4, 2, 5, 3))(x)
        x = Reshape((c, H, W))(x)
        x = Permute((2, 3, 1))(x)
        return x
    return layer 

def build_model(state_size, num_actions):
    input_shape = (4,) + state_size
    model = Sequential()
    if K.image_dim_ordering() == 'tf':
        # (width, height, channels)
        model.add(Permute((2, 3, 1), input_shape=input_shape))
    elif K.image_dim_ordering() == 'th':
        # (channels, width, height)
        model.add(Permute((1, 2, 3), input_shape=input_shape))
    else:
        raise RuntimeError('Unknown image_dim_ordering.')
    model.add(Convolution2D(32, 8, 8, subsample=(4, 4)))
    model.add(Activation('relu'))
    model.add(Convolution2D(64, 4, 4, subsample=(2, 2)))
    model.add(Activation('relu'))
    model.add(Convolution2D(64, 3, 3, subsample=(1, 1)))
    model.add(Activation('relu'))
    model.add(Flatten())
    model.add(Dense(512))
    model.add(Activation('relu'))
    model.add(Dense(num_actions))
    model.add(Activation('linear'))
    print(model.summary())
    return model 

def build_model(state_size, num_actions):
    input_shape = (4,) + state_size
    model = Sequential()
    if K.image_dim_ordering() == 'tf':
        # (width, height, channels)
        model.add(Permute((2, 3, 1), input_shape=input_shape))
    elif K.image_dim_ordering() == 'th':
        # (channels, width, height)
        model.add(Permute((1, 2, 3), input_shape=input_shape))
    else:
        raise RuntimeError('Unknown image_dim_ordering.')
    model.add(Convolution2D(32, 8, 8, subsample=(4, 4)))
    model.add(Activation('relu'))
    model.add(Convolution2D(64, 4, 4, subsample=(2, 2)))
    model.add(Activation('relu'))
    model.add(Convolution2D(64, 3, 3, subsample=(1, 1)))
    model.add(Activation('relu'))
    model.add(Flatten())
    model.add(Dense(512))
    model.add(Activation('relu'))
    model.add(Dense(num_actions))
    model.add(Activation('linear'))
    print(model.summary())
    return model 

def DUC(factor=(8, 8)):

    if factor[0] != factor[1]:
        raise ValueError('DUC upconvolution support only equal factors, '
                         'got {}'.format(factor))
    factor = factor[0]

    def layer(input_tensor):

        h, w, c = int_shape(input_tensor)[1:]
        H = h * factor
        W = w * factor

        x = Conv2DBlock(c*factor**2, (1,1),
                        padding='same',
                        name='duc_{}'.format(factor))(input_tensor)
        x = Permute((3, 1, 2))(x)
        x = Reshape((c, factor, factor, h, w))(x)
        x = Permute((1, 4, 2, 5, 3))(x)
        x = Reshape((c, H, W))(x)
        x = Permute((2, 3, 1))(x)
        return x
    return layer 

def create_Kao_Onet( weight_path = 'model48.h5'):
    input = Input(shape = [48,48,3])
    x = Conv2D(32, (3, 3), strides=1, padding='valid', name='conv1')(input)
    x = PReLU(shared_axes=[1,2],name='prelu1')(x)
    x = MaxPool2D(pool_size=3, strides=2, padding='same')(x)
    x = Conv2D(64, (3, 3), strides=1, padding='valid', name='conv2')(x)
    x = PReLU(shared_axes=[1,2],name='prelu2')(x)
    x = MaxPool2D(pool_size=3, strides=2)(x)
    x = Conv2D(64, (3, 3), strides=1, padding='valid', name='conv3')(x)
    x = PReLU(shared_axes=[1,2],name='prelu3')(x)
    x = MaxPool2D(pool_size=2)(x)
    x = Conv2D(128, (2, 2), strides=1, padding='valid', name='conv4')(x)
    x = PReLU(shared_axes=[1,2],name='prelu4')(x)
    x = Permute((3,2,1))(x)
    x = Flatten()(x)
    x = Dense(256, name='conv5') (x)
    x = PReLU(name='prelu5')(x)

    classifier = Dense(2, activation='softmax',name='conv6-1')(x)
    bbox_regress = Dense(4,name='conv6-2')(x)
    landmark_regress = Dense(10,name='conv6-3')(x)
    model = Model([input], [classifier, bbox_regress, landmark_regress])
    model.load_weights(weight_path, by_name=True)

    return model 

def create_Kao_Rnet (weight_path = 'model24.h5'):
    input = Input(shape=[24, 24, 3])  # change this shape to [None,None,3] to enable arbitraty shape input
    x = Conv2D(28, (3, 3), strides=1, padding='valid', name='conv1')(input)
    x = PReLU(shared_axes=[1, 2], name='prelu1')(x)
    x = MaxPool2D(pool_size=3,strides=2, padding='same')(x)

    x = Conv2D(48, (3, 3), strides=1, padding='valid', name='conv2')(x)
    x = PReLU(shared_axes=[1, 2], name='prelu2')(x)
    x = MaxPool2D(pool_size=3, strides=2)(x)

    x = Conv2D(64, (2, 2), strides=1, padding='valid', name='conv3')(x)
    x = PReLU(shared_axes=[1, 2], name='prelu3')(x)
    x = Permute((3, 2, 1))(x)
    x = Flatten()(x)
    x = Dense(128, name='conv4')(x)
    x = PReLU( name='prelu4')(x)
    classifier = Dense(2, activation='softmax', name='conv5-1')(x)
    bbox_regress = Dense(4, name='conv5-2')(x)
    model = Model([input], [classifier, bbox_regress])
    model.load_weights(weight_path, by_name=True)
    return model 

def stateless_attention_model(**kwargs):
    X = LSTM(kwargs['hidden_units'], kernel_initializer='he_normal', activation='tanh',
             dropout=kwargs['dropout'], return_sequences=True)(kwargs['embeddings'])
    attention_layer = Permute((2, 1))(X)
    attention_layer = Dense(kwargs['max_tweet_length'], activation='softmax')(attention_layer)
    attention_layer = Lambda(lambda x: K.mean(x, axis=1), name='dim_reduction')(attention_layer)
    attention_layer = RepeatVector(int(X.shape[2]))(attention_layer)
    attention_probabilities = Permute((2, 1), name='attention_probs')(attention_layer)
    attention_layer = Multiply()([X, attention_probabilities])
    attention_layer = Flatten()(attention_layer)
    return attention_layer 

def attention(inputs, single_attention_vector=False):
    # attention机制
    time_steps = k_keras.int_shape(inputs)[1]
    input_dim = k_keras.int_shape(inputs)[2]
    x = Permute((2, 1))(inputs)
    x = Dense(time_steps, activation='softmax')(x)
    if single_attention_vector:
        x = Lambda(lambda x: k_keras.mean(x, axis=1))(x)
        x = RepeatVector(input_dim)(x)

    a_probs = Permute((2, 1))(x)
    output_attention_mul = Multiply()([inputs, a_probs])
    return output_attention_mul 

def attention(inputs, single_attention_vector=False):
    # attention机制
    time_steps = k_keras.int_shape(inputs)[1]
    input_dim = k_keras.int_shape(inputs)[2]
    x = Permute((2, 1))(inputs)
    x = Dense(time_steps, activation='softmax')(x)
    if single_attention_vector:
        x = Lambda(lambda x: k_keras.mean(x, axis=1))(x)
        x = RepeatVector(input_dim)(x)

    a_probs = Permute((2, 1))(x)
    output_attention_mul = Multiply()([inputs, a_probs])
    return output_attention_mul 

def soft_attention_alignment(input_1, input_2):
    "Align text representation with neural soft attention"
    attention = Dot(axes=-1)([input_1, input_2])
    w_att_1 = Lambda(lambda x: softmax(x, axis=1),
                     output_shape=unchanged_shape)(attention)
    w_att_2 = Permute((2, 1))(Lambda(lambda x: softmax(x, axis=2),
                             output_shape=unchanged_shape)(attention))
    in1_aligned = Dot(axes=1)([w_att_1, input_1])
    in2_aligned = Dot(axes=1)([w_att_2, input_2])
    return in1_aligned, in2_aligned 

def create_model(input_shape, config):


    input_tensor = Input(shape=input_shape)  # this assumes K.image_dim_ordering() == 'tf'
    inception_model = InceptionV3(include_top=False, weights=None, input_tensor=input_tensor)
    # inception_model.load_weights("logs/2016-12-18-13-56-44/weights.21.model", by_name=True)

    for layer in inception_model.layers:
        layer.trainable = False

    x = inception_model.output
    #x = GlobalAveragePooling2D()(x)

    # (bs, y, x, c) --> (bs, x, y, c)
    x = Permute((2, 1, 3))(x)

    # (bs, x, y, c) --> (bs, x, y * c)
    _x, _y, _c = [int(s) for s in x._shape[1:]]
    x = Reshape((_x, _y*_c))(x)
    x = Bidirectional(LSTM(512, return_sequences=False), merge_mode="concat")(x)

    predictions = Dense(config["num_classes"], activation='softmax')(x)

    model = Model(input=inception_model.input, output=predictions)
    model.load_weights("logs/2017-01-02-13-39-41/weights.06.model")

    return model 

def create_Rnet(weight_path):
    input = Input(shape=[24, 24, 3])
    # 24,24,3 -> 11,11,28
    x = Conv2D(28, (3, 3), strides=1, padding='valid', name='conv1')(input)
    x = PReLU(shared_axes=[1, 2], name='prelu1')(x)
    x = MaxPool2D(pool_size=3,strides=2, padding='same')(x)

    # 11,11,28 -> 4,4,48
    x = Conv2D(48, (3, 3), strides=1, padding='valid', name='conv2')(x)
    x = PReLU(shared_axes=[1, 2], name='prelu2')(x)
    x = MaxPool2D(pool_size=3, strides=2)(x)

    # 4,4,48 -> 3,3,64
    x = Conv2D(64, (2, 2), strides=1, padding='valid', name='conv3')(x)
    x = PReLU(shared_axes=[1, 2], name='prelu3')(x)
    # 3,3,64 -> 64,3,3
    x = Permute((3, 2, 1))(x)
    x = Flatten()(x)
    # 576 -> 128
    x = Dense(128, name='conv4')(x)
    x = PReLU( name='prelu4')(x)
    # 128 -> 2 128 -> 4
    classifier = Dense(2, activation='softmax', name='conv5-1')(x)
    bbox_regress = Dense(4, name='conv5-2')(x)
    model = Model([input], [classifier, bbox_regress])
    model.load_weights(weight_path, by_name=True)
    return model

#-----------------------------#
#   mtcnn的第三段
#   精修框并获得五个点
#-----------------------------# 

def create_Onet(weight_path):
    input = Input(shape = [48,48,3])
    # 48,48,3 -> 23,23,32
    x = Conv2D(32, (3, 3), strides=1, padding='valid', name='conv1')(input)
    x = PReLU(shared_axes=[1,2],name='prelu1')(x)
    x = MaxPool2D(pool_size=3, strides=2, padding='same')(x)
    # 23,23,32 -> 10,10,64
    x = Conv2D(64, (3, 3), strides=1, padding='valid', name='conv2')(x)
    x = PReLU(shared_axes=[1,2],name='prelu2')(x)
    x = MaxPool2D(pool_size=3, strides=2)(x)
    # 8,8,64 -> 4,4,64
    x = Conv2D(64, (3, 3), strides=1, padding='valid', name='conv3')(x)
    x = PReLU(shared_axes=[1,2],name='prelu3')(x)
    x = MaxPool2D(pool_size=2)(x)
    # 4,4,64 -> 3,3,128
    x = Conv2D(128, (2, 2), strides=1, padding='valid', name='conv4')(x)
    x = PReLU(shared_axes=[1,2],name='prelu4')(x)
    # 3,3,128 -> 128,12,12
    x = Permute((3,2,1))(x)

    # 1152 -> 256
    x = Flatten()(x)
    x = Dense(256, name='conv5') (x)
    x = PReLU(name='prelu5')(x)

    # 鉴别
    # 256 -> 2 256 -> 4 256 -> 10 
    classifier = Dense(2, activation='softmax',name='conv6-1')(x)
    bbox_regress = Dense(4,name='conv6-2')(x)
    landmark_regress = Dense(10,name='conv6-3')(x)

    model = Model([input], [classifier, bbox_regress, landmark_regress])
    model.load_weights(weight_path, by_name=True)

    return model 

def permute(layer, layer_in, layerId, tensor=True):
    out = {layerId: Permute(map(int, layer['params']['dim'].split(',')))}
    if tensor:
        out[layerId] = out[layerId](*layer_in)
    return out 

def test_keras_import(self):
        model = Sequential()
        model.add(Permute((2, 1), input_shape=(64, 10)))
        model.build()
        self.keras_type_test(model, 0, 'Permute') 

def test_keras_export(self):
        tests = open(os.path.join(settings.BASE_DIR, 'tests', 'unit', 'keras_app',
                                  'keras_export_test.json'), 'r')
        response = json.load(tests)
        tests.close()
        net = yaml.safe_load(json.dumps(response['net']))
        net = {'l0': net['Input2'], 'l1': net['Permute']}
        net['l0']['connection']['output'].append('l1')
        inp = data(net['l0'], '', 'l0')['l0']
        net = permute(net['l1'], [inp], 'l1')
        model = Model(inp, net['l1'])
        self.assertEqual(model.layers[1].__class__.__name__, 'Permute') 

def model_CNN(p, embedding_matrix, max_sent_len, n_out):
    print("Parameters:", p)
    # Take sentence encoded as indices split in three parts and convert it to embeddings
    sentence_input = layers.Input(shape=(max_sent_len,), dtype='int32', name='sentence_input')
    word_embeddings = layers.Embedding(output_dim=embedding_matrix.shape[1],
                                       input_dim=embedding_matrix.shape[0],
                                       input_length=max_sent_len, weights=[embedding_matrix],
                                       mask_zero=True, trainable=False)(sentence_input)
    word_embeddings = layers.Dropout(p['dropout1'])(word_embeddings)

    # Take token markers that identify entity positions, convert to position embeddings
    entity_markers = layers.Input(shape=(2, max_sent_len,), dtype='int8', name='entity_markers')

    pos_embeddings = layers.wrappers.TimeDistributed(layers.Embedding(output_dim=p['position_emb'], input_dim=(max_sent_len*2)+1, input_length=max_sent_len,
                                                                      mask_zero=False, embeddings_regularizer = regularizers.l2(), trainable=True),  name='pos_embedding')(entity_markers)

    pos_embeddings = layers.Permute((2,1,3))(pos_embeddings)
    pos_embeddings = layers.Reshape((max_sent_len, p['position_emb']*2))(pos_embeddings)

    # Merge word and position embeddings and apply the specified amount of CNN layers
    x = layers.concatenate([word_embeddings, pos_embeddings])

    x = MaskedConvolution1D(nb_filter=p['units1'], filter_length=p['window_size'], border_mode='same')(x)
    sentence_vector = MaskedGlobalMaxPooling1D()(x)

    sentence_vector = layers.Lambda(lambda l: K.tanh(l))(sentence_vector)

    # Apply softmax
    sentence_vector = layers.Dropout(p['dropout1'])(sentence_vector)
    main_output = layers.Dense(n_out, activation="softmax", name='main_output')(sentence_vector)

    model = models.Model(input=[sentence_input, entity_markers], output=[main_output])
    model.compile(optimizer=p['optimizer'], loss='categorical_crossentropy', metrics=['accuracy'])

    return model 

def spatial_attention(input_feature):
	kernel_size = 7
	
	if K.image_data_format() == "channels_first":
		channel = input_feature._keras_shape[1]
		cbam_feature = Permute((2,3,1))(input_feature)
	else:
		channel = input_feature._keras_shape[-1]
		cbam_feature = input_feature
	
	avg_pool = Lambda(lambda x: K.mean(x, axis=3, keepdims=True))(cbam_feature)
	assert avg_pool._keras_shape[-1] == 1
	max_pool = Lambda(lambda x: K.max(x, axis=3, keepdims=True))(cbam_feature)
	assert max_pool._keras_shape[-1] == 1
	concat = Concatenate(axis=3)([avg_pool, max_pool])
	assert concat._keras_shape[-1] == 2
	cbam_feature = Conv2D(filters = 1,
					kernel_size=kernel_size,
					strides=1,
					padding='same',
					activation='sigmoid',
					kernel_initializer='he_normal',
					use_bias=False)(concat)	
	assert cbam_feature._keras_shape[-1] == 1
	
	if K.image_data_format() == "channels_first":
		cbam_feature = Permute((3, 1, 2))(cbam_feature)
		
	return multiply([input_feature, cbam_feature]) 

def attention_temporal(self, input_data, sequence_length):
        """
        A temporal attention layer
        :param input_data: Network input
        :param sequence_length: Length of the input sequence
        :return: The output of attention layer
        """
        a = Permute((2, 1))(input_data)
        a = Dense(sequence_length, activation='sigmoid')(a)
        a_probs = Permute((2, 1))(a)
        output_attention_mul = Multiply()([input_data, a_probs])
        return output_attention_mul 

def duc(x, factor=8, output_shape=(512, 512, 1)):
    if K.image_data_format() == 'channels_last':
        bn_axis = 3
    else:
        bn_axis = 1
    H, W, c, r = output_shape[0], output_shape[1], output_shape[2], factor
    h = H / r
    w = W / r
    x = Conv2D(
            c*r*r,
            (3, 3),
            padding='same',
            name='conv_duc_%s'%factor)(x)
    x = BatchNormalization(axis=bn_axis,name='bn_duc_%s'%factor)(x)
    x = Activation('relu')(x)
    x = Permute((3, 1, 2))(x)
    x = Reshape((c, r, r, h, w))(x)
    x = Permute((1, 4, 2, 5, 3))(x)
    x = Reshape((c, H, W))(x)
    x = Permute((2, 3, 1))(x)

    return x


# interpolation 

def squeeze_excite_block(input, ratio=16):
    ''' Create a channel-wise squeeze-excite block

    Args:
        input: input tensor
        filters: number of output filters

    Returns: a keras tensor

    References
    -   [Squeeze and Excitation Networks](https://arxiv.org/abs/1709.01507)
    '''
    init = input
    channel_axis = 1 if K.image_data_format() == "channels_first" else -1
    filters = init._keras_shape[channel_axis]
    se_shape = (1, 1, filters)

    se = GlobalAveragePooling2D()(init)
    se = Reshape(se_shape)(se)
    se = Dense(filters // ratio, activation='relu', kernel_initializer='he_normal', use_bias=False)(se)
    se = Dense(filters, activation='sigmoid', kernel_initializer='he_normal', use_bias=False)(se)

    if K.image_data_format() == 'channels_first':
        se = Permute((3, 1, 2))(se)

    x = multiply([init, se])
    return x 

def squeeze_excite_block(input_tensor, ratio=16):
    """ Create a channel-wise squeeze-excite block

    Args:
        input_tensor: input Keras tensor
        ratio: number of output filters

    Returns: a Keras tensor

    References
    -   [Squeeze and Excitation Networks](https://arxiv.org/abs/1709.01507)
    """
    init = input_tensor
    channel_axis = 1 if K.image_data_format() == "channels_first" else -1
    filters = _tensor_shape(init)[channel_axis]
    se_shape = (1, 1, filters)

    se = GlobalAveragePooling2D()(init)
    se = Reshape(se_shape)(se)
    se = Dense(filters // ratio, activation='relu', kernel_initializer='he_normal', use_bias=False)(se)
    se = Dense(filters, activation='sigmoid', kernel_initializer='he_normal', use_bias=False)(se)

    if K.image_data_format() == 'channels_first':
        se = Permute((3, 1, 2))(se)

    x = multiply([init, se])
    return x