为什么tf.image.grayscale到rgb在tensorflow 1.14.0上不起作用？

Question

我正在尝试将MNIST数据集转换为RGB。我使用下面的代码来实现这一点，但是我得到了一个错误，它适用于tensorflow1.12.0，但不适用于tensorflow1.14.0（我的项目基于tensorflow1.14.0）

mnist = tf.keras.datasets.mnist
(x_train, _), (x_test, _) = mnist.load_data()

X = np.concatenate([x_train, x_test])
X = X.reshape((70000, 28, 28, 1))
assert X.shape == (70000, 28, 28, 1)

X = tf.image.grayscale_to_rgb(X, name=None)
assert X.shape == (70000, 28, 28, 3)

我得到的错误是：

Traceback (most recent call last):
  File "PGE/redis_ako.py", line 75, in <module>
    X2 = tf.image.grayscale_to_rgb(X, name=None)
  File "/home/farshad/.virtualenvs/ako_cifar/local/lib/python2.7/site-packages/tensorflow/python/ops/image_ops_impl.py", line 1827, in grayscale_to_rgb
    images = _AssertGrayscaleImage(images)
  File "/home/farshad/.virtualenvs/ako_cifar/local/lib/python2.7/site-packages/tensorflow/python/ops/image_ops_impl.py", line 263, in _AssertGrayscaleImage
    _CheckGrayscaleImage(image, require_static=False), image)
  File "/home/farshad/.virtualenvs/ako_cifar/local/lib/python2.7/site-packages/tensorflow/python/ops/image_ops_impl.py", line 282, in _CheckGrayscaleImage
    if image.get_shape().ndims is None:
AttributeError: 'numpy.ndarray' object has no attribute 'get_shape'