InvalidArgumentError: Несовместимые фигуры: [6291456] и [8388608] - PullRequest
Новая
       37

InvalidArgumentError: Несовместимые фигуры: [6291456] и [8388608]

0 голосов
/ 01 мая 2020

Я попытался запустить код модели keras: 

def new_model(input_shape=(4, 128, 128, 128), n_base_filters=16, depth=5, dropout_rate=0.3,
                      n_segmentation_levels=3, n_labels=4, optimizer=Adam, initial_learning_rate=5e-4,
                      loss_function=weighted_dice_coefficient_loss, activation_name="sigmoid",metrics=dice_coefficient):

    inputs = Input(input_shape)
    split = Lambda( lambda x: tf.split(x,num_or_size_splits=4,axis=1))(inputs)

    p_4 = tf.keras.layers.Conv3D(
        name               = "p_4",
        filters            = 1,
        kernel_size        = (1,1,1),
        strides            = (1,1,1),
        padding            = "valid",
        dilation_rate      = (1,1,1),
        use_bias           = True,
        kernel_initializer = "glorot_uniform",
        bias_initializer   = "zeros",
    )(split[0])
    p_5 = tf.keras.layers.Conv3D(
        name               = "p_5",
        filters            = 1,
        kernel_size        = (1,1,1),
        strides            = (1,1,1),
        padding            = "valid",
        dilation_rate      = (1,1,1),
        use_bias           = True,
        kernel_initializer = "glorot_uniform",
        bias_initializer   = "zeros",
    )(split[1])
    p_6 = tf.keras.layers.Conv3D(
        name               = "p_6",
        filters            = 1,
        kernel_size        = (1,1,1),
        strides            = (1,1,1),
        padding            = "valid",
        dilation_rate      = (1,1,1),
        use_bias           = True,
        kernel_initializer = "glorot_uniform",
        bias_initializer   = "zeros",
    )(split[2])
    p_7 = tf.keras.layers.Conv3D(
        name               = "p_7",
        filters            = 1,
        kernel_size        = (1,1,1),
        strides            = (1,1,1),
        padding            = "valid",
        dilation_rate      = (1,1,1),
        use_bias           = True,
        kernel_initializer = "glorot_uniform",
        bias_initializer   = "zeros",
    )(split[3])
    p_8 = tf.keras.layers.Activation(
        name       = "p_8",
        activation = activation_name,
    )(
        tf.keras.layers.Concatenate(name='concat_p_4_p_5_p_6_p_7', axis=0)([
            p_4,
            p_5,
            p_6,
            p_7
        ])
    )


    output_layer = p_8


    model = Model(inputs=inputs, outputs=output_layer)

    if not isinstance(metrics, list):
        metrics = [metrics]
#     model.compile(optimizer=optimizer(lr=initial_learning_rate), loss=loss_function)        
    model.compile(optimizer=optimizer(lr=initial_learning_rate), loss=loss_function, metrics=metrics)
    return model

Когда я запустил эту модель, я получил следующую ошибку. Я сомневаюсь, что проблема вызвана из-за лямбда-оператора. 

Invalid ArgumentError:  Incompatible shapes: [6291456] vs. [8388608]
     [[node mul (defined at ../unet3d/metrics.py:9) ]] [Op:__inference_train_function_921]

Errors may have originated from an input operation.
Input Source operations connected to node mul:
 Reshape (defined at /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:2700)

Function call stack:
train_function
...