def test_hyper_parameters_guided_style_loss(subtests, impl_params):
hyper_parameters = paper.hyper_parameters(impl_params=impl_params)
sub_params = "guided_style_loss"
assert sub_params in hyper_parameters
parameters = getattr(hyper_parameters, sub_params)
layers, num_channels = zip(
("relu1_1" if impl_params else "conv1_1", 64),
("relu2_1" if impl_params else "conv2_1", 128),
("relu3_1" if impl_params else "conv3_1", 256),
("relu4_1" if impl_params else "conv4_1", 512),
("relu5_1" if impl_params else "conv5_1", 512),
)
layer_weights = [1 / n**2 for n in num_channels]
with subtests.test("layers"):
assert parameters.layers == layers
with subtests.test("layer_weights"):
assert parameters.layer_weights == pytest.approx(layer_weights)
with subtests.test("region_weights"):
assert parameters.region_weights == "sum"
with subtests.test("score_weight"):
assert parameters.score_weight == pytest.approx(1e3)
def test_content_loss(subtests):
content_loss = paper.content_loss()
assert isinstance(content_loss, pystiche.loss.FeatureReconstructionLoss)
hyper_parameters = paper.hyper_parameters().content_loss
with subtests.test("layer"):
assert content_loss.encoder.layer == hyper_parameters.layer
with subtests.test("score_weight"):
assert content_loss.score_weight == pytest.approx(hyper_parameters.score_weight)
def test_hyper_parameters_content_loss(subtests, impl_params):
hyper_parameters = paper.hyper_parameters(impl_params=impl_params)
sub_params = "content_loss"
assert sub_params in hyper_parameters
hyper_parameters = getattr(hyper_parameters, sub_params)
with subtests.test("layer"):
assert hyper_parameters.layer == "relu4_2" if impl_params else "conv4_2"
with subtests.test("score_weight"):
assert hyper_parameters.score_weight == pytest.approx(1e0)
def test_image_pyramid(subtests):
image_pyramid = paper.image_pyramid()
assert isinstance(image_pyramid, pyramid.ImagePyramid)
levels = tuple(iter(image_pyramid))
assert len(levels) == 2
edge_sizes, num_steps = zip(*[(level.edge_size, level.num_steps)
for level in levels])
hyper_parameters = paper.hyper_parameters().image_pyramid
with subtests.test("edge_sizes"):
assert edge_sizes == hyper_parameters.edge_sizes
with subtests.test("num_steps"):
assert num_steps == hyper_parameters.num_steps
def test_hyper_parameters_image_pyramid(subtests, impl_params):
hyper_parameters = paper.hyper_parameters(impl_params=impl_params)
sub_params = "image_pyramid"
assert sub_params in hyper_parameters
hyper_parameters = getattr(hyper_parameters, sub_params)
with subtests.test("edge_sizes"):
assert hyper_parameters.edge_sizes == (512 if impl_params else 500,
1024)
with subtests.test("num_steps"):
if impl_params:
assert hyper_parameters.num_steps == (500, 200)
else:
assert len(hyper_parameters.num_steps) == 2
ratio = hyper_parameters.num_steps[0] / hyper_parameters.num_steps[
1]
assert ratio == pytest.approx(2.5)
def test_style_loss(subtests):
style_loss = paper.style_loss()
assert isinstance(style_loss, pystiche.loss.MultiLayerEncodingLoss)
hyper_parameters = paper.hyper_parameters().style_loss
with subtests.test("losses"):
assert all(
isinstance(loss, pystiche.loss.GramLoss) for loss in style_loss.children()
)
layers, layer_weights = zip(
*[(loss.encoder.layer, loss.score_weight) for loss in style_loss.children()]
)
with subtests.test("layers"):
assert layers == hyper_parameters.layers
with subtests.test("layer_weights"):
assert layer_weights == pytest.approx(hyper_parameters.layer_weights)
with subtests.test("score_weight"):
assert style_loss.score_weight == pytest.approx(hyper_parameters.score_weight)
def test_hyper_parameters_smoke(impl_params):
hyper_parameters = paper.hyper_parameters(impl_params=impl_params)
assert isinstance(hyper_parameters, HyperParameters)
