def test_film(self, resnet_size, enabled_blocks):
image = tf.zeros((2, 224, 224, 3), dtype=tf.float32)
embedding = tf.zeros((2, 100), dtype=tf.float32)
film_generator_fn = functools.partial(
resnet.linear_film_generator, enabled_block_layers=enabled_blocks)
_ = resnet.resnet_model(image,
is_training=True,
num_classes=1001,
resnet_size=resnet_size,
return_intermediate_values=True,
film_generator_fn=film_generator_fn,
film_generator_input=embedding)
def test_intermediate_values(self, scope):
with tf.variable_scope(scope):
image = tf.zeros((2, 224, 224, 3), dtype=tf.float32)
end_points = resnet.resnet_model(image,
is_training=True,
num_classes=1001,
return_intermediate_values=True)
tensors = ['initial_conv', 'initial_max_pool', 'pre_final_pool',
'final_reduce_mean', 'final_dense']
tensors += [
'block_layer{}'.format(i + 1) for i in range(4)]
self.assertEqual(set(tensors), set(end_points.keys()))
def test_malformed_film_raises(self):
image = tf.zeros((2, 224, 224, 3), dtype=tf.float32)
embedding = tf.zeros((2, 100), dtype=tf.float32)
film_generator_fn = functools.partial(resnet.linear_film_generator,
enabled_block_layers=[True] * 5)
with self.assertRaises(ValueError):
_ = resnet.resnet_model(image,
is_training=True,
num_classes=1001,
resnet_size=18,
return_intermediate_values=True,
film_generator_fn=film_generator_fn,
film_generator_input=embedding)
