def _input_and_output_same_shape_helper(self, kernel_size):
img_batch = tf.placeholder(tf.float32, shape=[None, 32, 32, 3])
output_img_batch, _ = cyclegan.cyclegan_generator_resnet(
img_batch, kernel_size=kernel_size)
self.assertAllEqual(img_batch.shape.as_list(),
output_img_batch.shape.as_list())
def test_generator_inference(self):
"""Check one inference step."""
img_batch = tf.zeros([2, 32, 32, 3])
model_output, _ = cyclegan.cyclegan_generator_resnet(img_batch)
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
sess.run(model_output)
def generator(input_images):
"""Thin wrapper around CycleGAN generator to conform to the TFGAN API.
Args:
input_images: A batch of images to translate. Images should be normalized
already. Shape is [batch, height, width, channels].
Returns:
Returns generated image batch.
Raises:
ValueError: If shape of last dimension (channels) is not defined.
"""
input_images.shape.assert_has_rank(4)
input_size = input_images.shape.as_list()
channels = input_size[-1]
if channels is None:
raise ValueError('Last dimension shape must be known but is None: %s' %
input_size)
with tf.contrib.framework.arg_scope(cyclegan.cyclegan_arg_scope()):
output_images, _ = cyclegan.cyclegan_generator_resnet(
input_images, num_outputs=channels)
return output_images
def test_generator_unknown_batch_dim(self):
"""Check that generator can take unknown batch dimension inputs."""
img = tf.placeholder(tf.float32, shape=[None, 32, None, 3])
output_imgs, _ = cyclegan.cyclegan_generator_resnet(img)
self.assertAllEqual([None, 32, None, 3], output_imgs.shape.as_list())
def _test_generator_graph_helper(self, shape):
"""Check that generator can take small and non-square inputs."""
output_imgs, _ = cyclegan.cyclegan_generator_resnet(tf.ones(shape))
self.assertAllEqual(shape, output_imgs.shape.as_list())