From c4fe8e5009283f4ab539b3a6d52af7deaf2b1c81 Mon Sep 17 00:00:00 2001 From: lmj Date: Mon, 19 Sep 2022 19:35:18 +0800 Subject: [PATCH] update MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 删除文件 examples/face_adversarial_attack/example update update examples/face_adversarial_attack/example/AFR.py. Signed-off-by: 君君臣臣君 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 update examples/face_adversarial_attack/example/AFR.py. 修改了命名错误；包的顺序。 Signed-off-by: 君君臣臣君 update examples/face_adversarial_attack/example/example_non-target_attack.py. 修改包的名称。 Signed-off-by: 君君臣臣君 update examples/face_adversarial_attack/example/example_target_attack.py. Signed-off-by: 君君臣臣君 update examples/face_adversarial_attack/example/loss_design.py. Signed-off-by: 君君臣臣君 update examples/face_adversarial_attack/example/loss_design.py. Signed-off-by: 君君臣臣君 --- .../face_adversarial_attack/example/AFR.py | 67 +++++++----------- .../example/example_non-target_attack.py | 2 +- .../example/example_target_attack.py | 2 +- .../example/loss_design.py | 68 ++++++------------- .../face_adversarial_attack/example/test.py | 2 +- 5 files changed, 47 insertions(+), 94 deletions(-) diff --git a/examples/face_adversarial_attack/example/AFR.py b/examples/face_adversarial_attack/example/AFR.py index 3b64290..7c1fde4 100644 --- a/examples/face_adversarial_attack/example/AFR.py +++ b/examples/face_adversarial_attack/example/AFR.py @@ -15,25 +15,25 @@ import os import re import numpy as np -import matplotlib.image as mp +import mindspore +import mindspore.dataset.vision.py_transforms as P +from mindspore.dataset.vision.py_transforms import ToPIL as ToPILImage +from mindspore import Parameter, ops, nn, Tensor +from mindspore.dataset.vision.py_transforms import ToTensor import dlib +import matplotlib.image as mp import face_recognition as fr import face_recognition_models as frm from PIL import Image, ImageDraw -import mindspore -from mindspore.dataset.vision.py_transforms import ToPIL as ToPILImage -from mindspore import Parameter,ops, nn,Tensor -from mindspore.dataset.vision.py_transforms import ToTensor -import mindspore.dataset.vision.py_transforms as P -from loss_design import TrainOneStepCell,MyWithLossCell,FaceLoss_no_target_attack,FaceLoss_target_attack +from loss_design import MyTrainOneStepCell, MyWithLossCell, FaceLossTargetAttack, FaceLossNoTargetAttack from FaceRecognition.eval import get_net -class Attack(object): +class FaceAdversarialAttack(object): """ Class used to create adversarial facial recognition attacks """ - def __init__(self,input_img,target_img,seed=None): + def __init__(self, input_img, target_img, seed=None): """ Initialization for Attack class. @@ -44,7 +44,7 @@ class Attack(object): """ - if (seed != None): np.random.seed(seed) + if (seed is not None): np.random.seed(seed) self.MEAN = Tensor([0.485, 0.456, 0.406]) self.STD = Tensor([0.229, 0.224, 0.225]) self.LOSS = Tensor(0) @@ -56,11 +56,12 @@ class Attack(object): self.input_tensor = Tensor(self.normalize(self.tensorize(input_img))) self.target_tensor = Tensor(self.normalize(self.tensorize(target_img))) mp.imsave('./outputs/input图像.jpg', np.transpose(self._reverse_norm(self.input_tensor).asnumpy(), (1, 2, 0))) - mp.imsave('./outputs/target图像.jpg', np.transpose(self._reverse_norm(self.target_tensor).asnumpy(), (1, 2, 0))) + mp.imsave('./outputs/target图像.jpg', + np.transpose(self._reverse_norm(self.target_tensor).asnumpy(), (1, 2, 0))) - self.input_emb = self.resnet(self.expand_dims(self.input_tensor,0)) - self.target_emb = self.resnet(self.expand_dims(self.target_tensor,0)) + self.input_emb = self.resnet(self.expand_dims(self.input_tensor, 0)) + self.target_emb = self.resnet(self.expand_dims(self.target_tensor, 0)) self.adversarial_emb = None self.mask_tensor = self._create_mask(input_img) self.ref = self.mask_tensor @@ -69,28 +70,29 @@ class Attack(object): - def train(self,attack_method): + def train(self, attack_method): """ Optimized adversarial image. """ if attack_method == "non-target attack": - LOSS = FaceLoss_no_target_attack(self.target_emb) + LOSS = FaceLossNoTargetAttack(self.target_emb) if attack_method == "target_attack": - LOSS = FaceLoss_target_attack(self.target_emb) + LOSS = FaceLossTargetAttack(self.target_emb) - net_with_criterion = MyWithLossCell(self.resnet, LOSS,self.input_tensor) - train_net = TrainOneStepCell(net_with_criterion, self.opt) + net_with_criterion = MyWithLossCell(self.resnet, LOSS, self.input_tensor) + train_net = MyTrainOneStepCell(net_with_criterion, self.opt) for i in range(2000): self.mask_tensor = Tensor(self.pm) - grads,loss = train_net(self.mask_tensor) + grads, loss = train_net(self.mask_tensor) print("epoch %d ,loss: %f \n " % (i, loss.asnumpy().item())) - self.mask_tensor = ops.clip_by_value(self.mask_tensor, Tensor(0, mindspore.float32), Tensor(1, mindspore.float32)) + self.mask_tensor = ops.clip_by_value( + self.mask_tensor, Tensor(0, mindspore.float32), Tensor(1, mindspore.float32)) adversarial_tensor = self._apply( self.input_tensor, @@ -124,33 +126,12 @@ class Attack(object): print("================================") print("adversarial:", adversarial) print("adversarial_confidence:", self.adversarial_emb.asnumpy()[0][adversarial]) - print("Confidence changes for target::", self.adversarial_emb.asnumpy()[0][target]) - print("Confidence changes for input:", self.adversarial_emb.asnumpy()[0][input]) - print("================================") - print("target:", target) - print("target_confidence:", self.target_emb.asnumpy()[0][target]) - print("input:%d, target:%d, adversarial:%d" % (input,target, adversarial)) - - def test1(self,adversarial_tensor): - self.adversarial_emb = self.resnet(self.expand_dims((adversarial_tensor - self.MEAN[:, None, None]) / self.STD[:, None, None], 0)) - self.input_emb = self.resnet(self.expand_dims(self.input_tensor, 0)) - self.target_emb = self.resnet(self.expand_dims(self.target_tensor, 0)) - - adversarial = np.argmax(self.adversarial_emb.asnumpy()) - target = np.argmax(self.target_emb.asnumpy()) - input = np.argmax(self.input_emb.asnumpy()) - - print("input:", input) - print("input_confidence:", self.input_emb.asnumpy()[0][input]) - print("================================") - print("adversarial:", adversarial) - print("adversarial_confidence:", self.adversarial_emb.asnumpy()[0][adversarial]) + print("Confidence changes for target:", self.adversarial_emb.asnumpy()[0][target]) print("Confidence changes for input:", self.adversarial_emb.asnumpy()[0][input]) print("================================") print("target:", target) print("target_confidence:", self.target_emb.asnumpy()[0][target]) - print("input:%d, target:%d, adversarial:%d" % (input, target, adversarial)) - + print("input: %d, target: %d, adversarial: %d" % (input, target, adversarial)) def _reverse_norm(self, image_tensor): diff --git a/examples/face_adversarial_attack/example/example_non-target_attack.py b/examples/face_adversarial_attack/example/example_non-target_attack.py index a1fd4e2..42d70e2 100644 --- a/examples/face_adversarial_attack/example/example_non-target_attack.py +++ b/examples/face_adversarial_attack/example/example_non-target_attack.py @@ -29,7 +29,7 @@ if __name__ == '__main__': targets = AFR.load_data('opencv_photo/target/') - adversarial = AFR.Attack(inputs[0], targets[0]) + adversarial = AFR.FaceAdversarialAttack(inputs[0], targets[0]) attack_method = "non-target attack" adversarial_tensor, mask_tensor = adversarial.train(attack_method) diff --git a/examples/face_adversarial_attack/example/example_target_attack.py b/examples/face_adversarial_attack/example/example_target_attack.py index 37fbead..7e089dd 100644 --- a/examples/face_adversarial_attack/example/example_target_attack.py +++ b/examples/face_adversarial_attack/example/example_target_attack.py @@ -29,7 +29,7 @@ if __name__ == '__main__': targets = AFR.load_data('opencv_photo/target/') - adversarial = AFR.Attack(inputs[0], targets[0]) + adversarial = AFR.FaceAdversarialAttack(inputs[0], targets[0]) attack_method = "target_attack" diff --git a/examples/face_adversarial_attack/example/loss_design.py b/examples/face_adversarial_attack/example/loss_design.py index df8015c..2e70955 100644 --- a/examples/face_adversarial_attack/example/loss_design.py +++ b/examples/face_adversarial_attack/example/loss_design.py @@ -13,14 +13,14 @@ # limitations under the License. # ============================================================================ import mindspore -from mindspore import ops, nn,Tensor +from mindspore import ops, nn, Tensor from mindspore.dataset.vision.py_transforms import ToTensor import mindspore.dataset.vision.py_transforms as P -class TrainOneStepCell(nn.Cell): +class MyTrainOneStepCell(nn.TrainOneStepCell): """ Encapsulation class of network training. @@ -34,24 +34,22 @@ 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): + def construct(self, *inputs): """Defines the computation performed.""" loss = self.network(*inputs) grads = self.grad(self.network)(*inputs) self.optimizer(grads) - return grads,loss + return grads, loss + + class MyWithLossCell(nn.Cell): - def __init__(self,net,loss_fn,input_tensor): + def __init__(self, net, loss_fn, input_tensor): super(MyWithLossCell, self).__init__(auto_prefix=False) self.net = net self._loss_fn = loss_fn @@ -63,11 +61,14 @@ class MyWithLossCell(nn.Cell): self.input_tensor = input_tensor self.input_emb = self.net(self.expand_dims(self.input_tensor, 0)) - def construct(self,mask_tensor): + def construct(self, mask_tensor): ref = mask_tensor - adversarial_tensor = mindspore.numpy.where((ref == 0), self.input_tensor, (mask_tensor - self.MEAN[:, None, None] )/ self.STD[:, None, None]) + adversarial_tensor = mindspore.numpy.where( + (ref == 0), + self.input_tensor, + (mask_tensor - self.MEAN[:, None, None] )/ self.STD[:, None, None]) adversarial_emb = self.net(self.expand_dims(adversarial_tensor, 0)) - loss = self._loss_fn( adversarial_emb,self.input_emb,mask_tensor) + loss = self._loss_fn( adversarial_emb, self.input_emb, mask_tensor) return loss @property @@ -75,11 +76,11 @@ class MyWithLossCell(nn.Cell): return self.net -class FaceLoss_target_attack(nn.Cell): +class FaceLossTargetAttack(nn.Cell): """The loss function of the target attack""" - def __init__(self,target_emb): - super(FaceLoss_target_attack, self).__init__() + def __init__(self, target_emb): + super(FaceLossTargetAttack, self).__init__() self.uniformreal = ops.UniformReal(seed=2) self.sum = ops.ReduceSum(keep_dims=False) self.norm = nn.Norm(keep_dims=True) @@ -92,21 +93,7 @@ class FaceLoss_target_attack(nn.Cell): self.abs = ops.Abs() 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) + def construct(self, adversarial_emb, input_emb, mask_tensor): 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,)) @@ -116,11 +103,11 @@ class FaceLoss_target_attack(nn.Cell): return loss -class FaceLoss_no_target_attack(nn.Cell): +class FaceLossNoTargetAttack(nn.Cell): """The loss function of the non-target attack""" def __init__(self, target_emb): """初始化""" - super(FaceLoss_no_target_attack, self).__init__() + super(FaceLossNoTargetAttack, self).__init__() self.uniformreal = ops.UniformReal(seed=2) self.sum = ops.ReduceSum(keep_dims=False) self.norm = nn.Norm(keep_dims=True) @@ -134,21 +121,6 @@ class FaceLoss_no_target_attack(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,)) diff --git a/examples/face_adversarial_attack/example/test.py b/examples/face_adversarial_attack/example/test.py index 95d3a0b..7bbc62b 100644 --- a/examples/face_adversarial_attack/example/test.py +++ b/examples/face_adversarial_attack/example/test.py @@ -15,7 +15,7 @@ import numpy as np import matplotlib.image as mp -from mindspore import context,Tensor +from mindspore import context, Tensor import mindspore from mindspore.dataset.vision.py_transforms import ToTensor import mindspore.dataset.vision.py_transforms as P