update examples/face_adversarial_attack/example/loss_design.py.

MyTrainOneStepCell继承了nn.TrainOneStepCell减少了代码量；删除了无用代码；修改了命名错误。

Signed-off-by: 君君臣臣君 <mingjun@isrc.iscas.ac.cn>

examples/face_adversarial_attack/example/loss_design.py

@@ -20,7 +20,7 @@ import mindspore.dataset.vision.py_transforms as P
 
 
 
-class TrainOneStepCell(nn.Cell):
+class MyTrainOneStepCell(nn.TrainOneStepCell):
     """
     Encapsulation class of network training.
 
@@ -34,11 +34,7 @@ class TrainOneStepCell(nn.Cell):
     """
 
     def __init__(self, network, optimizer, sens=1.0):
-        super(TrainOneStepCell, self).__init__(auto_prefix=False)
-        self.network = network
-        self.network.set_grad()
-        self.optimizer = optimizer
-        self.weights = self.optimizer.parameters
+        super(MyTrainOneStepCell, self).__init__(network, optimizer,sens)
         self.grad = ops.composite.GradOperation(get_all=True, sens_param=False)
 
     def construct(self, *inputs):
@@ -47,6 +43,8 @@ class TrainOneStepCell(nn.Cell):
         grads = self.grad(self.network)(*inputs)
         self.optimizer(grads)
         return grads, loss
+    
+    
 
 
 
@@ -78,7 +76,7 @@ class MyWithLossCell(nn.Cell):
        return self.net
 
 
-class FaceLossTargeTattack(nn.Cell):
+class FaceLossTargetAttack(nn.Cell):
     """The loss function of the target attack"""
 
     def __init__(self, target_emb):
@@ -96,20 +94,6 @@ class FaceLossTargeTattack(nn.Cell):
         self.reduce_mean = ops.ReduceMean()
 
     def construct(self, adversarial_emb, input_emb, mask_tensor):
-        #像素平滑
-        # vert_diff = mask_tensor[:, 1:] - mask_tensor[:, :-1]
-        # hor_diff = mask_tensor[:, :, 1:] - mask_tensor[:, :, :-1]
-        # vert_diff_sq = self.pow(vert_diff, 2)
-        # hor_diff_sq = self.pow(hor_diff, 2)
-        # A = self.zeroslike(Tensor(self.uniformreal((3, 1, 112))))
-        # B = self.zeroslike(Tensor(self.uniformreal((3, 112, 1))))
-        # vert_pad = self.concat_op1((vert_diff_sq, A))
-        # hor_pad = self.concat_op2((hor_diff_sq, B))
-        # tv_sum = vert_pad + hor_pad
-        # tv = ops.functional.sqrt(tv_sum + 1e-5)
-        # tv_final_sum = self.sum(tv)
-        # tv_loss = (1e-4) * tv_final_sum
-        # print("tv_loss:",tv_loss)
         prod_sum = self.reduce_sum(adversarial_emb * self.target_emb, (1,))
         square1 = self.reduce_sum(ops.functional.square(adversarial_emb), (1,))
         square2 = self.reduce_sum(ops.functional.square(self.target_emb), (1,))
@@ -137,21 +121,6 @@ class FaceLossNoTargetAttack(nn.Cell):
         self.reduce_mean = ops.ReduceMean()
 
     def construct(self, adversarial_emb, input_emb, mask_tensor):
-        # 像素平滑
-        # vert_diff = mask_tensor[:, 1:] - mask_tensor[:, :-1]  #
-        # hor_diff = mask_tensor[:, :, 1:] - mask_tensor[:, :, :-1]
-        # vert_diff_sq = self.pow(vert_diff, 2)
-        # hor_diff_sq = self.pow(hor_diff, 2)
-        # A = self.zeroslike(Tensor(self.uniformreal((3, 1, 112))))  #
-        # B = self.zeroslike(Tensor(self.uniformreal((3, 112, 1))))
-        # vert_pad = self.concat_op1((vert_diff_sq, A))
-        # hor_pad = self.concat_op2((hor_diff_sq, B))
-        # tv_sum = vert_pad + hor_pad
-        # tv = ops.functional.sqrt(tv_sum + 1e-5)
-        # tv_final_sum = self.sum(tv)
-        # tv_loss = (1e-4) * tv_final_sum
-        # print("tv_loss:",tv_loss)
-
         prod_sum = self.reduce_sum(adversarial_emb * input_emb, (1,))
         square1 = self.reduce_sum(ops.functional.square(adversarial_emb), (1,))
         square2 = self.reduce_sum(ops.functional.square(input_emb), (1,))