tensorflow.python.framework.errors_impl.OperatorNotAllowedInGraphError

前提

https://github.com/IDEALLab/bezier-gan
上記のURL「Bézier-GAN」を実行しようとしています。
その際に今回のエラーが発生しました。

実現したいこと

tensorflow.python.framework.errors_impl.OperatorNotAllowedInGraphErrorを改善したい

発生している問題・エラーメッセージ

エラーメッセージ
Plotting training samples ...
Traceback (most recent call last):
  File "train_gan.py", line 60, in <module>
    model.train(X_train, batch_size=batch_size, train_steps=train_steps, save_interval=args.save_interval, directory=directory)
  File "C:\Users\phanton\Desktop\?\bezier-gan-master\bezier-gan-master\beziergan\gan.py", line 253, in train
    d_real, _ = self.discriminator(self.x)
  File "C:\Users\phanton\Desktop\?\bezier-gan-master\bezier-gan-master\beziergan\gan.py", line 166, in discriminator
    padding='same')
  File "C:\Users\phanton\Desktop\?\pycharm_env\python3.7\lib\site-packages\keras\dtensor\utils.py", line 96, in _wrap_function
    init_method(layer_instance, *args, **kwargs)
  File "C:\Users\phanton\Desktop\?\pycharm_env\python3.7\lib\site-packages\keras\layers\convolutional\conv2d.py", line 197, in __init__
    **kwargs
  File "C:\Users\phanton\Desktop\?\pycharm_env\python3.7\lib\site-packages\keras\layers\convolutional\base_conv.py", line 128, in __init__
    if filters is not None and filters <= 0:
  File "C:\Users\phanton\Desktop\?\pycharm_env\python3.7\lib\site-packages\tensorflow\python\framework\ops.py", line 955, in __bool__
    self._disallow_bool_casting()
  File "C:\Users\phanton\Desktop\?\pycharm_env\python3.7\lib\site-packages\tensorflow\python\framework\ops.py", line 559, in _disallow_bool_casting
    "Using a symbolic `tf.Tensor` as a Python `bool`")
  File "C:\Users\phanton\Desktop\?\pycharm_env\python3.7\lib\site-packages\tensorflow\python\framework\ops.py", line 543, in _disallow_in_graph_mode
    f"{task} is not allowed in Graph execution. Use Eager execution or"
tensorflow.python.framework.errors_impl.OperatorNotAllowedInGraphError: Using a symbolic `tf.Tensor` as a Python `bool` is not allowed in Graph execution. Use Eager execution or decorate this function with @tf.function.

該当のソースコード

python
1    def discriminator(self, x, reuse=tf.AUTO_REUSE, training=True):
2
3        depth = 64
4        dropout = 0.4
5        kernel_size = (4, 2)
6
7        with tf.variable_scope('Discriminator', reuse=reuse):
8            x = tf.keras.layers.Conv2D(x, (depth * 1, kernel_size),
9                                       strides=(2, 1),
10                                       padding='same')
11            x = tf.keras.layers.BatchNormalization(x,
12                                                   momentum=0.9)  # , training=training)
13            x = tf.nn.leaky_relu(x, alpha=0.2)
14            x = tf.keras.layers.Dropout(x, dropout, training=training)
15
16            x = tf.keras.layers.Conv2D(x, (depth * 2, kernel_size),
17                                       strides=(2, 1),
18                                       padding='same')
19            x = tf.keras.layers.BatchNormalization(x,
20                                                   momentum=0.9)  # , training=training)
21            x = tf.nn.leaky_relu(x, alpha=0.2)
22            x = tf.keras.layers.dropout(x, dropout, training=training)
23
24            x = tf.keras.layers.Conv2D(x, (depth * 4, kernel_size),
25                                       strides=(2, 1),
26                                       padding='same')
27            x = tf.keras.layers.BatchNormalization(x,
28                                                   momentum=0.9)  # , training=training)
29            x = tf.nn.leaky_relu(x, alpha=0.2)
30            x = tf.keras.layers.dropout(x, dropout, training=training)
31
32            x = tf.keras.layers.Conv2D(x, (depth * 8, kernel_size),
33                                       strides=(2, 1),
34                                       padding='same')
35            x = tf.keras.layers.BatchNormalization(x,
36                                                   momentum=0.9)  # , training=training)
37            x = tf.nn.leaky_relu(x, alpha=0.2)
38            x = tf.keras.layers.dropout(x, dropout, training=training)
39
40            x = tf.keras.layers.Conv2D(x, (depth * 16, kernel_size),
41                                       strides=(2, 1),
42                                       padding='same')
43            x = tf.keras.layers.BatchNormalization(x,
44                                                   momentum=0.9)  # , training=training)
45            x = tf.nn.leaky_relu(x, alpha=0.2)
46            x = tf.keras.layers.dropout(x, dropout, training=training)
47
48            x = tf.keras.layers.Conv2D(x, (depth * 32, kernel_size),
49                                       strides=(2, 1),
50                                       padding='same')
51            x = tf.keras.layers.BatchNormalization(x,
52                                                   momentum=0.9)  # , training=training)
53            x = tf.nn.leaky_relu(x, alpha=0.2)
54            x = tf.keras.layers.dropout(x, dropout, training=training)
55
56            x = tf.keras.layers.flatten(x)
57            x = tf.keras.layers.dense(x, 1024)
58            x = tf.keras.layers.BatchNormalization(x,
59                                                   momentum=0.9)  # , training=training)
60            x = tf.nn.leaky_relu(x, alpha=0.2)
61
62            d = tf.keras.layers.dense(x, 1)
63
64            q = tf.keras.layers.dense(x, 128)
65            q = tf.keras.layers.BatchNormalization(q,
66                                                   momentum=0.9)  # , training=training)
67            q = tf.nn.leaky_relu(q, alpha=0.2)
68            q_mean = tf.keras.layers.Dense(q, self.latent_dim)
69            q_logstd = tf.keras.layers.Dense(q, self.latent_dim)
70            q_logstd = tf.maximum(q_logstd, -16)
71            # Reshape to batch_size x 1 x latent_dim
72            q_mean = tf.reshape(q_mean, (-1, 1, self.latent_dim))
73            q_logstd = tf.reshape(q_logstd, (-1, 1, self.latent_dim))
74            q = tf.concat([q_mean, q_logstd], axis=1,
75                          name='predicted_latent')  # batch_size x 2 x latent_dim
76
77            return d, q

keras\dtensor\utils.py.line96
1 def _wrap_function(layer_instance, *args, **kwargs):
2        signature = inspect.signature(init_method)
3        layout_args = {}
4        # Check args like 'kernel_initializer' and pop the 'kernel_layout' if it
5        # presents.
6        for variable_name in KERAS_VARIABLE_NAMES:
7            if variable_name + "_initializer" in signature.parameters:
8                layout = kwargs.pop(variable_name + "_layout", None)
9                if layout:
10                    layout_args[variable_name + "_layout"] = layout
11
12        init_method(layer_instance, *args, **kwargs)
13
14        # Inject the layout parameter after the invocation of __init__()
15        for layout_param_name, layout in layout_args.items():
16            setattr(layer_instance, layout_param_name, layout)
17
18    # return decorated
19    return tf.__internal__.decorator.make_decorator(
20        target=init_method, decorator_func=_wrap_function
21    )

試したこと

以下のURLを参考に
https://stackoverflow.com/questions/64698542/typeerror-init-got-multiple-values-for-argument-strides
x = tf.keras.layers.Conv2D(x, (depth * 1, kernel_size),
strides=(2, 1),
padding='same')
とかっこをつけてみたのですが
tensorflow.python.framework.errors_impl.OperatorNotAllowedInGraphError: Using a symbolic tf.Tensor as a Python bool is not allowed in Graph execution. Use Eager execution or decorate this function with @tf.function.
とエラーが出てしまいました。