#
# mask = torch.tensor([0, 1, 0, 1, 1, 0], device='cuda')
# source = torch.tensor([1.0, 2.0, 3.0], device='cuda')
# mask = torch.tensor([0, 1, 0, 1, 1, 0], device='cuda', dtype=torch.uint8)
# y = myfunc(mask, source)
# example 2
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
self.layer = torch.nn.Linear(2, 1)
def forward(self, x):
return self.layer(x)
x = torch.tensor([[0.0, 0.0], [0.0, 1.0], [1.0, 0.0], [1.0, 1.0]])
y = torch.tensor([3.0, 5.0, 4.0, 6.0])
model = Model()
optimizer = torch.optim.SGD(model.parameters(), lr=0.1)
with torchsnooper.snoop():
for _ in range(100):
optimizer.zero_grad()
# pred = model(x)
pred = model(x).squeeze()
squared_diff = (y - pred) ** 2
loss = squared_diff.mean()
print(loss.item())
loss.backward()
optimizer.step()
'''
'''
#no-targeted L2 attack
adversary = L2MomentumIterativeAttack(
learner.model, loss_fn=attack_cosine_distance(target=torch.ones(1).to(conf.device)), eps=10,
nb_iter=40, eps_iter=0.2, decay_factor=1., clip_min=-1.0, clip_max=1.0,
targeted=False)
'''
#image = Image.open("/hd1/anshengnan/InsightFace_Pytorch/data/test.jpg")
loader, class_num = get_train_loader(conf)
head = Arcface(embedding_size=conf.embedding_size,
classnum=class_num).to(conf.device)
with ts.snoop():
for imgs, labels in tqdm(iter(loader)):
imgs = imgs.to(conf.device)
labels = labels.to(conf.device)
embeddings = learner.model(imgs)
thetas = head(embeddings, labels)
adv_targeted = adversary.perturb(imgs, labels)
adv_embeddings = learner.model(adv_targeted)
adv_thetas = head(adv_embeddings, labels)
print(labels)
thetas = list(thetas.squeeze(0))
adv_thetas = list(adv_thetas.squeeze(0))
print(thetas.index(max(thetas)))
print(adv_thetas.index(max(adv_thetas)))
break