Lambda 层：函数构建代码之外的操作正在传递一个“图形”张量

Question

我有一个类似这个帖子的问题：Tensorflow 2.1.0 - An op outside of the function building code is being passed a "Graph" tensor

我的模型包括一个围绕 CNN 的 TimeDistributed 包装器，其输入层首先由 lambda 函数进行归一化：

    # ENCODER
    # video input
    frames = 10
    video_input = tf.keras.Input(shape=(frames, 256, 256, 3))

    # cnn 1 frame input
    frame_in = Input(shape=(256,256,3), name='frame_input')
    #normalisation layer
    cnn = Lambda(lambda inputs: frame_in / 255)(frame_in)
    # add cnn layers + activations
    cnn = Conv2D(filters=num_filters[0], kernel_size=(4,4), strides=(4,4), padding='same', activation = 'relu', data_format='channels_last', name = 'Layer_1')(cnn) #input_shape=(64,64,3), data_format='channels_last'), name = 'Layer_1')(frame_in)
    cnn = Conv2D(filters=num_filters[1], kernel_size=(4,4), strides=(4,4), padding='same', activation = 'relu', name = 'Layer_2')(cnn)
    cnn = Conv2D(filters=num_filters[2], kernel_size=(4,4), strides=(2,2), padding='same', activation = 'relu', name = 'Layer_3')(cnn)
    cnn = Conv2D(filters=num_filters[3], kernel_size=(4,4), strides=(2,2), padding='same', activation = 'relu', name = 'Layer_4')(cnn)
    cnn = Conv2D(filters=num_filters[4], kernel_size=(4,4), strides=(1,1), padding='valid', activation = 'relu', name = 'Layer_5')(cnn)
    # flatten
    cnn = Flatten()(cnn)
    #time distributed
    time_dist_cnn = Model(inputs=frame_in, outputs=cnn, name='Time_Distributed_CNN')
    time_dist_0 = TimeDistributed(time_dist_cnn)(video_input)

错误是一样的：

TypeError: An op outside of the function building code is being passed
a "Graph" tensor. It is possible to have Graph tensors
leak out of the function building context by including a
tf.init_scope in your function building code.
For example, the following function will fail:
  @tf.function
  def has_init_scope():
    my_constant = tf.constant(1.)
    with tf.init_scope():
      added = my_constant * 2
The graph tensor has name: frame_input:0

我知道问题几乎相同，并且源于使用 lambda 表达式，但我不确定如何在这种情况下解决它。

Answer 1

如果在 Eager 模式下运行，TensorFlow 操作将检查输入是否为 tensorflow.python.framework.ops.EagerTensor 类型，并且 keras 操作被实现为 DAG。因此，如果 Eager 模式的输入是 tensorflow.python.framework.ops.Tensor，则会引发错误。

您可以通过使用tf.config.experimental_run_functions_eagerly(True) 明确告诉 tensorflow 在 keras 的渴望模式下运行来将输入类型更改为 EagerTensor。添加此语句应该可以解决您的问题。

另外，通过在程序顶部添加tf.compat.v1.disable_eager_execution() 来禁用急切执行也会成功运行程序。

例如，这个程序抛出你所面临的错误 -

重现错误的代码-

import numpy as np
import tensorflow as tf
print(tf.__version__)
from tensorflow.keras import layers, losses, models

def get_loss_fcn(w):
    def loss_fcn(y_true, y_pred):
        loss = w * losses.mse(y_true, y_pred)
        return loss
    return loss_fcn

data_x = np.random.rand(5, 4, 1)
data_w = np.random.rand(5, 4)
data_y = np.random.rand(5, 4, 1)

x = layers.Input([4, 1])
w = layers.Input([4])
y = layers.Activation('tanh')(x)
model = models.Model(inputs=[x, w], outputs=y)
loss = get_loss_fcn(model.input[1])

model.compile(loss=loss)
model.fit((data_x, data_w), data_y)

输出 -

2.2.0
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/execute.py in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
     59     tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
---> 60                                         inputs, attrs, num_outputs)
     61   except core._NotOkStatusException as e:

TypeError: An op outside of the function building code is being passed
a "Graph" tensor. It is possible to have Graph tensors
leak out of the function building context by including a
tf.init_scope in your function building code.
For example, the following function will fail:
  @tf.function
  def has_init_scope():
    my_constant = tf.constant(1.)
    with tf.init_scope():
      added = my_constant * 2
The graph tensor has name: input_8:0

During handling of the above exception, another exception occurred:

_SymbolicException                        Traceback (most recent call last)
8 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/execute.py in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
     72       raise core._SymbolicException(
     73           "Inputs to eager execution function cannot be Keras symbolic "
---> 74           "tensors, but found {}".format(keras_symbolic_tensors))
     75     raise e
     76   # pylint: enable=protected-access

_SymbolicException: Inputs to eager execution function cannot be Keras symbolic tensors, but found [<tf.Tensor 'input_8:0' shape=(None, 4) dtype=float32>]

解决方案 1 - 在程序顶部添加此 tf.config.experimental_run_functions_eagerly(True) 会成功运行程序。在程序顶部添加tf.compat.v1.disable_eager_execution() 以禁用急切执行也会成功运行程序。

固定代码 -

import numpy as np
import tensorflow as tf
print(tf.__version__)
from tensorflow.keras import layers, losses, models

tf.config.experimental_run_functions_eagerly(True)

def get_loss_fcn(w):
    def loss_fcn(y_true, y_pred):
        loss = w * losses.mse(y_true, y_pred)
        return loss
    return loss_fcn

data_x = np.random.rand(5, 4, 1)
data_w = np.random.rand(5, 4)
data_y = np.random.rand(5, 4, 1)

x = layers.Input([4, 1])
w = layers.Input([4])
y = layers.Activation('tanh')(x)
model = models.Model(inputs=[x, w], outputs=y)
loss = get_loss_fcn(model.input[1])

model.compile(loss=loss)
model.fit((data_x, data_w), data_y)

print('Done.')

输出 -

2.2.0
1/1 [==============================] - 0s 1ms/step - loss: 0.0000e+00
Done.

---

解决方案2 -也请尝试以下解决方案，您可以尝试如下修改，看看是否有效 -

1) 创建normalize_input函数如下：

def normalize_input(inputs):
    norm_input = inputs / 255
    return norm_input

2) 修改Lambda层如下：

cnn = tf.keras.layers.Lambda(function=normalize_input)([frame_in])

希望这能回答您的问题。快乐学习。