def forward(self,
x,
target=None,
mixup=False,
mixup_hidden=False,
args=None,
grad=None,
noise=None,
adv_mask1=0,
adv_mask2=0,
mp=None):
if mixup_hidden:
layer_mix = random.randint(0, 2)
elif mixup:
layer_mix = 0
else:
layer_mix = None
out = x
if target is not None:
target_reweighted = to_one_hot(target, self.num_classes)
if layer_mix == 0:
out, target_reweighted = mixup_process(out,
target_reweighted,
args=args,
grad=grad,
noise=noise,
adv_mask1=adv_mask1,
adv_mask2=adv_mask2,
mp=mp)
out = self.conv1(out)
out = self.layer1(out)
if layer_mix == 1:
out, target_reweighted = mixup_process(out, target_reweighted, args=args, hidden=True)
out = self.layer2(out)
if layer_mix == 2:
out, target_reweighted = mixup_process(out, target_reweighted, args=args, hidden=True)
out = self.layer3(out)
if layer_mix == 3:
out, target_reweighted = mixup_process(out, target_reweighted, args=args, hidden=True)
out = act(self.bn1(out))
out = F.avg_pool2d(out, 8)
out = out.reshape(out.size(0), -1)
out = self.linear(out)
if target is not None:
return out, target_reweighted
else:
return out
def forward(self, x, target= None, mixup=False, mixup_hidden=False, args=None, grad=None,
noise=None, adv_mask1=0, adv_mask2=0, profile=None):
if mixup_hidden:
layer_mix = random.randint(0,2)
elif mixup:
layer_mix = 0
else:
layer_mix = None
out = x
if target is not None :
target_reweighted = to_one_hot(target,self.num_classes)
end =time.time()
if layer_mix == 0:
out, target_reweighted = mixup_process(out, target_reweighted, args=args, grad=grad, noise=noise, adv_mask1=adv_mask1, adv_mask2=adv_mask2)
profile['gc'].append(time.time()-end)
if len(profile['data'])%10 == 0: print('gc : ', np.mean(profile['gc']))
out = self.conv1(out)
out = self.layer1(out)
if layer_mix == 1:
out, target_reweighted = mixup_process(out, target_reweighted, args=args, hidden=True)
out = self.layer2(out)
if layer_mix == 2:
out, target_reweighted = mixup_process(out, target_reweighted, args=args, hidden=True)
out = self.layer3(out)
if layer_mix == 3:
out, target_reweighted = mixup_process(out, target_reweighted, args=args, hidden=True)
out = self.layer4(out)
out = F.avg_pool2d(out, 4)
out = out.view(out.size(0), -1)
out = self.linear(out)
if target is not None:
return out, target_reweighted
else:
return out
def forward(self,
x,
target=None,
mixup=False,
mixup_hidden=False,
args=None,
grad=None,
noise=None,
adv_mask1=0,
adv_mask2=0):
if mixup_hidden:
layer_mix = random.randint(0, 2)
elif mixup:
layer_mix = 0
else:
layer_mix = None
out = x
if target is not None:
target_reweighted = to_one_hot(target, self.num_classes)
if layer_mix == 0:
out, target_reweighted = mixup_process(out,
target_reweighted,
args=args,
grad=grad,
noise=noise,
adv_mask1=adv_mask1,
adv_mask2=adv_mask2)
out = self.conv1(out)
out = self.layer1(out)
if layer_mix == 1:
out, target_reweighted = mixup_process(out,
target_reweighted,
args=args,
hidden=True)
out = self.layer2(out)
if layer_mix == 2:
out, target_reweighted = mixup_process(out,
target_reweighted,
args=args,
hidden=True)
out = self.layer3(out)
if layer_mix == 3:
out, target_reweighted = mixup_process(out,
target_reweighted,
args=args,
hidden=True)
out = self.layer4(out)
out = F.avg_pool2d(out, out.shape[-1])
out = out.view(out.size(0), -1)
if self.weight_type is not None:
out = F.normalize(out, p=2, dim=1)
out = self.linear(out)
if target is not None:
return out, target_reweighted
else:
return out