如何为 LocallyConnected2D 层使用 WeightNormalization 包装器

Question

我正在尝试在 tf.layers.LocallyConnected2D 层周围使用 tfa.layers.WeightNormalization 包装器，如下所示：

from tensorflow_addons.layers import WeightNormalization
import tensorflow as tf

x = tf.ones((1, 32, 32, 3))
x = WeightNormalization(tf.keras.layers.LocallyConnected2D(3, 3))(x)

它给出了以下错误：

TypeError: 'NoneType' object is not callable

为了记录，这确实适用于Conv2D 层。知道如何使用LocallyConnected2D 层来实现这一点吗？

Answer 1

来自 cmets

from tensorflow_addons.layers import WeightNormalization
import tensorflow as tf

x = tf.ones((1, 32, 32, 3))
x = WeightNormalization(tf.keras.layers.LocallyConnected2D(3, 3), data_init=False)(x)

（转述自 Marco Cerliani）