def test_simple(self):
# channels_last
with self.test_session():
# create simple UpsampleLike layer
upsample_like_layer = layers.UpsampleLike()
# create input source
source = np.zeros((1, 2, 2, 1), dtype=K.floatx())
source = K.variable(source)
target = np.zeros((1, 5, 5, 1), dtype=K.floatx())
expected = target
target = K.variable(target)
# compute output
computed_shape = upsample_like_layer.compute_output_shape(
[source.shape, target.shape])
actual = upsample_like_layer.call([source, target])
actual = K.get_value(actual)
self.assertEqual(actual.shape, computed_shape)
self.assertAllEqual(actual, expected)
# channels_first
with self.test_session():
# create simple UpsampleLike layer
upsample_like_layer = layers.UpsampleLike(
data_format='channels_first')
# create input source
source = np.zeros((1, 1, 2, 2), dtype=K.floatx())
source = K.variable(source)
target = np.zeros((1, 1, 5, 5), dtype=K.floatx())
expected = target
target = K.variable(target)
# compute output
computed_shape = upsample_like_layer.compute_output_shape(
[source.shape, target.shape])
actual = upsample_like_layer.call([source, target])
actual = K.get_value(actual)
self.assertEqual(actual.shape, computed_shape)
self.assertAllEqual(actual, expected)
def test_mini_batch(self):
# create simple UpsampleLike layer
upsample_like_layer = layers.UpsampleLike()
# create input source
source = np.zeros((2, 2, 2, 1), dtype=K.floatx())
source = K.variable(source)
target = np.zeros((2, 5, 5, 1), dtype=K.floatx())
expected = target
target = K.variable(target)
# compute output
actual = upsample_like_layer.call([source, target])
actual = K.get_value(actual)
self.assertAllEqual(actual, expected)