Ошибка NoneType в лямбда-слое при наложении маски на тензор

Question

У меня есть набор данных X двухмерных изображений, и я разделил его на партии X_1, ..., X_k. Затем для каждого i-го компонента элемента пакета я хочу умножить его на параметр gamma_i и добавить к ним beta_i.

Ошибка:

Input reshape: (4, 12, 12, 1)
Traceback (most recent call last):
  File "3.py", line 61, in <module>
    model = build_test((i, 12, 12, 1))
  File "3.py", line 49, in build_test
    x = Lambda(lambda tensor, mask: tf.math.multiply(tensor, tf.convert_to_tensor(mask, dtype=tf.float32)))([input_img, gamma])
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 634, in __call__
    outputs = call_fn(inputs, *args, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/keras/layers/core.py", line 785, in call
    return self.function(inputs, **arguments)
  File "3.py", line 49, in <lambda>
    x = Lambda(lambda tensor, mask: tf.math.multiply(tensor, tf.convert_to_tensor(mask, dtype=tf.float32)))([input_img, gamma])
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1087, in convert_to_tensor
    return convert_to_tensor_v2(value, dtype, preferred_dtype, name)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1145, in convert_to_tensor_v2
    as_ref=False)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1224, in internal_convert_to_tensor
    ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py", line 305, in _constant_tensor_conversion_function
    return constant(v, dtype=dtype, name=name)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py", line 246, in constant
    allow_broadcast=True)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/framework/constant_op.py", line 284, in _constant_impl
    allow_broadcast=allow_broadcast))
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/framework/tensor_util.py", line 454, in make_tensor_proto
    raise ValueError("None values not supported.")
ValueError: None values not supported.

Выможет воспроизвести ошибку, используя код:

import tensorflow as tf
from tensorflow.keras.layers import Conv2D, Input, BatchNormalization
from tensorflow.keras.layers import Lambda, Multiply, Add, Reshape
from tensorflow.keras.models import Model
import tensorflow.keras.backend as K
import numpy as np

def build_test(input_shape):
    input_img = Input(shape=input_shape)
    print('Input reshape:', input_shape)

    x = Lambda(lambda tensor, mask: tf.math.multiply(tensor, tf.convert_to_tensor(mask, dtype=tf.float32)))([input_img, gamma])
    print('multiply:', x.shape)
    x = Lambda(lambda tensor, mask: tf.math.add(tensor, tf.convert_to_tensor(mask, dtype=tf.float32)))([x, beta])
    print('add:', x.shape)

    test = Model(inputs=input_img, outputs=x)
    return test

batch_sizes = [4, 8]
for i in batch_sizes:
    gamma = np.arange(i)
    beta = np.arange(i)
    model = build_test((i, 12, 12, 1))

Однако, когда я использую следующий код, который косвенно решает мою цель, работает нормально:

import tensorflow as tf
from tensorflow.keras.layers import Conv2D, Input, BatchNormalization
from tensorflow.keras.layers import Lambda, Multiply, Add, Reshape
from tensorflow.keras.models import Model
import tensorflow.keras.backend as K
import numpy as np

def build_test(input_shape):
    input_img = Input(shape=input_shape)
    print('Input reshape:', input_shape)

    for batch in range(input_shape[0]):
        # x = tf.gather(x, batch)
        # print('tensor shape:', x.shape)  
        mask = np.empty(input_shape, dtype=np.float32)
        # for gamma
        mask.fill(1)
        mask[batch].fill(gamma[batch])
        x = Lambda(lambda tensor: tf.math.multiply(tensor, tf.convert_to_tensor(mask, dtype=tf.float32)))(input_img)
        print('multiply:', x.shape)
        # for beta
        mask.fill(1)
        mask[batch].fill(beta[batch])
        x = Lambda(lambda tensor: tf.math.add(tensor, tf.convert_to_tensor(mask, dtype=tf.float32)))(x)
        print('add:', x.shape)

    test = Model(inputs=input_img, outputs=x)
    return test

batch_sizes = [4, 8]
for i in batch_sizes:
    gamma = np.arange(i)
    beta = np.arange(i)
    model = build_test((i, 12, 12, 1))

Что я делаю неправильно в первомкод, для которого я получаю ошибку?

Answer 1

В первом коде при инициализации маска не была объявлена ​​ранее, и, следовательно, значение тензора для нее было пустым и выдавало ошибку.

Поскольку в последнем коде вы его инициализировали, он работал нормально.