def test_pso_attack():
"""
PSO_Attack test
"""
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
batch_size = 6
net = SimpleNet()
inputs = np.random.rand(batch_size, 10)
model = ModelToBeAttacked(net)
labels = np.random.randint(low=0, high=10, size=batch_size)
labels = np.eye(10)[labels]
labels = labels.astype(np.float32)
attack = PSOAttack(model, bounds=(0.0, 1.0), pm=0.5, sparse=False)
_, adv_data, _ = attack.generate(inputs, labels)
assert np.any(inputs != adv_data)
def test_pso_attack_detection_cpu():
"""
PSO_Attack test
"""
context.set_context(mode=context.GRAPH_MODE, device_target="CPU")
batch_size = 2
inputs = np.random.random((batch_size, 100, 100, 3))
model = DetectionModel()
attack = PSOAttack(model,
t_max=30,
pm=0.5,
model_type='detection',
reserve_ratio=0.5)
# generate adversarial samples
adv_imgs = []
for i in range(batch_size):
img_data = np.expand_dims(inputs[i], axis=0)
pre_gt_boxes, pre_gt_labels = model.predict(inputs)
_, adv_img, _ = attack.generate(img_data,
(pre_gt_boxes, pre_gt_labels))
adv_imgs.append(adv_img)
assert np.any(inputs != np.array(adv_imgs))