def test_preprocessing_defences(self):
from art.estimators.object_detection.pytorch_faster_rcnn import PyTorchFasterRCNN
from art.defences.preprocessor.spatial_smoothing import SpatialSmoothing
pre_def = SpatialSmoothing()
frcnn = PyTorchFasterRCNN(
clip_values=(0, 1),
attack_losses=("loss_classifier", "loss_box_reg",
"loss_objectness", "loss_rpn_box_reg"),
preprocessing_defences=pre_def,
)
# Create labels
result = frcnn.predict(x=self.x_test)
y = [
{
"boxes": result[0]["boxes"],
"labels": result[0]["labels"],
"scores": np.ones_like(result[0]["labels"]),
},
{
"boxes": result[1]["boxes"],
"labels": result[1]["labels"],
"scores": np.ones_like(result[1]["labels"]),
},
]
# Compute gradients
grads = frcnn.loss_gradient(x=self.x_test, y=y)
self.assertTrue(grads.shape == (2, 28, 28, 3))
def test_defence_non_pytorch(get_default_mnist_subset, image_dl_estimator,
device_type):
smooth_3x3 = SpatialSmoothing(window_size=3, channels_first=True)
preprocessing_defences = [smooth_3x3]
_test_preprocessing_defences_forward(get_default_mnist_subset,
image_dl_estimator, device_type,
preprocessing_defences)
_test_preprocessing_defences_backward(get_default_mnist_subset,
image_dl_estimator, device_type,
preprocessing_defences)
def test_defence_non_tensorflow(get_default_mnist_subset, image_dl_estimator,
is_tf_version_2):
if is_tf_version_2:
smooth_3x3 = SpatialSmoothing(window_size=3, channels_first=False)
preprocessing_defences = [smooth_3x3]
device_type = None
_test_preprocessing_defences_forward(get_default_mnist_subset,
image_dl_estimator, device_type,
preprocessing_defences)
_test_preprocessing_defences_backward(get_default_mnist_subset,
image_dl_estimator, device_type,
preprocessing_defences)
def test_defence_non_pytorch(art_warning, get_default_mnist_subset, image_dl_estimator, device_type):
try:
smooth_3x3 = SpatialSmoothing(window_size=3, channels_first=True)
preprocessing_defences = [smooth_3x3]
_test_preprocessing_defences_forward(
get_default_mnist_subset, image_dl_estimator, device_type, preprocessing_defences
)
_test_preprocessing_defences_backward(
get_default_mnist_subset, image_dl_estimator, device_type, preprocessing_defences
)
except ARTTestException as e:
art_warning(e)
def test_defences_tensorflow_and_nontensorflow(art_warning, get_default_mnist_subset, image_dl_estimator, device_type):
try:
smooth_3x3_nonpth = SpatialSmoothing(window_size=3, channels_first=False)
smooth_3x3_pth = SpatialSmoothingTensorFlowV2(window_size=3, channels_first=False)
preprocessing_defences = [smooth_3x3_nonpth, smooth_3x3_pth]
device_type = None
_test_preprocessing_defences_forward(
get_default_mnist_subset, image_dl_estimator, device_type, preprocessing_defences
)
_test_preprocessing_defences_backward(
get_default_mnist_subset, image_dl_estimator, device_type, preprocessing_defences
)
except ARTTestException as e:
art_warning(e)
def test_preprocessing_defences(art_warning, get_mnist_dataset):
try:
from art.estimators.object_detection.tensorflow_faster_rcnn import TensorFlowFasterRCNN
from art.defences.preprocessor.spatial_smoothing import SpatialSmoothing
images = tf.placeholder(tf.float32, shape=[1, 28, 28, 3])
pre_def = SpatialSmoothing()
with pytest.raises(ValueError):
_ = TensorFlowFasterRCNN(
images=images,
clip_values=(0, 1),
attack_losses=("loss_classifier", "loss_box_reg",
"loss_objectness", "loss_rpn_box_reg"),
preprocessing_defences=pre_def,
)
except ARTTestException as e:
art_warning(e)
def test_compute_loss(self):
from art.estimators.object_detection.pytorch_faster_rcnn import PyTorchFasterRCNN
from art.defences.preprocessor.spatial_smoothing import SpatialSmoothing
pre_def = SpatialSmoothing()
frcnn = PyTorchFasterRCNN(
clip_values=(0, 1),
attack_losses=("loss_classifier", "loss_box_reg",
"loss_objectness", "loss_rpn_box_reg"),
preprocessing_defences=pre_def,
)
# Create labels
result = frcnn.predict(
np.repeat(self.x_test_mnist[:2].astype(np.float32),
repeats=3,
axis=3))
y = [
{
"boxes": result[0]["boxes"],
"labels": result[0]["labels"],
"scores": np.ones_like(result[0]["labels"]),
},
{
"boxes": result[1]["boxes"],
"labels": result[1]["labels"],
"scores": np.ones_like(result[1]["labels"]),
},
]
# Compute loss
loss = frcnn.compute_loss(x=self.x_test, y=y)
self.assertAlmostEqual(float(loss), 0.6324392, delta=0.01)
