python - 如何为 LocallyConnected2D 层使用 WeightNormalization 包装器

Question

我正在尝试在这样的层tfa.layers.WeightNormalization周围使用包装器：tf.layers.LocallyConnected2D

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

作为记录，这确实适用于Conv2D图层。知道如何让这个与LocallyConnected2D图层一起工作吗？

标签： pythontensorflowkerasnormalizationkeras-layer

Answer 1

来自评论

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）