def test_upsample_value(self):
image = tf.random.normal([10, 32, 32, 3])
big_image = generator.usample(image)
expected_image = tf.compat.v1.image.resize_nearest_neighbor(
image, [32 * 2, 32 * 2])
with self.cached_session() as sess:
big_image_np, expected_image_np = sess.run([big_image, expected_image])
self.assertAllEqual(big_image_np, expected_image_np)
def test_usample_shapes_value_placeholder(self):
"""Tests usample with partially shaped inputs."""
if tf.executing_eagerly():
# tf.placeholder() is not compatible with eager execution.
return
image = tf.placeholder(tf.float32, shape=[None, 32, 32, 3])
big_image = generator.usample(image)
self.assertEqual([None, 64, 64, 3], big_image.shape.as_list())
expected_image = tf.image.resize_nearest_neighbor(
image, [32 * 2, 32 * 2])
with self.cached_session() as sess:
big_image_np, expected_image_np = sess.run(
[big_image, expected_image],
{image: np.random.normal(size=[10, 32, 32, 3])})
self.assertAllEqual(big_image_np, expected_image_np)
def test_usample_shapes(self):
"""Tests that upsampling has the desired effect on shape."""
image = tf.random.normal([10, 32, 32, 3])
big_image = generator.usample(image)
self.assertEqual([10, 64, 64, 3], big_image.shape.as_list())