diff --git a/examples/community/face_adversarial_attack/README.md b/examples/community/face_adversarial_attack/README.md index 5a8b1e6..396f22e 100644 --- a/examples/community/face_adversarial_attack/README.md +++ b/examples/community/face_adversarial_attack/README.md @@ -3,12 +3,12 @@ ##描述 -本项目是对人脸识别模型的物理对抗攻击，通过生成对抗口罩，使人脸佩戴后实现目标攻击和非目标攻击，并应用于mindspore平台。 +本项目是对人脸识别模型的物理对抗攻击，通过生成对抗口罩，使人脸佩戴后实现有目标攻击和非目标攻击，并应用于MindSpore平台。 ##模型结构 -采用华为mindspore官方训练的FaceRecognition模型 +采用华为MindSpore官方训练的FaceRecognition模型 https://www.mindspore.cn/resources/hub/details?MindSpore/1.7/facerecognition_ms1mv2 @@ -21,16 +21,15 @@ mindspore=1.7，硬件平台为GPU。 ##脚本说明 ├── readme.md -├── opencv_photo +├── photos │ ├── adv_input //对抗图像 │ ├── input //输入图像 │ └── target //目标图像 ├── outputs //训练后的图像 ├── FaceRecognition //模型设置 -├── AFR.py //训练脚本 -├── camera.py //opencv图像采集 +├── adversarial_attack.py //训练脚本 │── example_non-target_attack.py //无目标攻击训练 -│── example_target_attack.py //目标攻击训练 +│── example_target_attack.py //有目标攻击训练 │── loss_design.py //训练优化设置 └── test.py //评估攻击效果 @@ -58,14 +57,14 @@ mindspore=1.7，硬件平台为GPU。 利用 MindSpore代码仓中的https://gitee.com/mindspore/models/blob/master/research/cv/FaceRecognition/eval.py的get_model函数加载模型 ##训练过程 -目标攻击: +有目标攻击: - $ cd face_adversarial_attack/example/ + $ cd face_adversarial_attack/ $ python example_target_attack.py 非目标攻击: - $ cd face_adversarial_attack/example/ + $ cd face_adversarial_attack/ $ python example_non-target_attack.py @@ -78,9 +77,13 @@ optimizer=adam, learning rate=0.01, weight_decay=0.0001, epoch=2000 ##评估过程 评估方法一: - AFR.Attack.test() + adversarial_attack.Attack.test() + 评估方法二: - $ cd face_adversarial_attack/example/ + $ cd face_adversarial_attack/ $ python test.py + + + diff --git a/examples/community/face_adversarial_attack/AFR.py b/examples/community/face_adversarial_attack/adversarial_attack.py similarity index 84% rename from examples/community/face_adversarial_attack/AFR.py rename to examples/community/face_adversarial_attack/adversarial_attack.py index 7c1fde4..d960ed6 100644 --- a/examples/community/face_adversarial_attack/AFR.py +++ b/examples/community/face_adversarial_attack/adversarial_attack.py @@ -15,39 +15,38 @@ import os import re import numpy as np +import face_recognition as fr +import face_recognition_models as frm +import dlib +import matplotlib.image as mp +from PIL import Image, ImageDraw 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 +from mindspore import Parameter, ops, nn, Tensor from loss_design import MyTrainOneStepCell, MyWithLossCell, FaceLossTargetAttack, FaceLossNoTargetAttack from FaceRecognition.eval import get_net class FaceAdversarialAttack(object): """ - Class used to create adversarial facial recognition attacks + Class used to create adversarial facial recognition attacks. """ def __init__(self, input_img, target_img, seed=None): """ - Initialization for Attack class. + Initialization for attack class. Args: - input_img : Image to train on. - target_img : Image to target the adversarial attack against. - seed : optional Sets custom seed for reproducability. Default is generated randomly. + input_img (numpy.ndarray): The input image. + target_img (numpy.ndarray): The target image. + seed (int): optional Sets custom seed for reproducability. Default is generated randomly. """ 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) self.expand_dims = mindspore.ops.ExpandDims() self.imageize = ToPILImage() self.tensorize = ToTensor() @@ -55,8 +54,8 @@ class FaceAdversarialAttack(object): self.resnet = get_net() 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', + mp.imsave('./outputs/input_image.jpg', np.transpose(self._reverse_norm(self.input_tensor).asnumpy(), (1, 2, 0))) + mp.imsave('./outputs/target_image.jpg', np.transpose(self._reverse_norm(self.target_tensor).asnumpy(), (1, 2, 0))) @@ -73,14 +72,22 @@ class FaceAdversarialAttack(object): def train(self, attack_method): """ Optimized adversarial image. + + Args: + attack_method (Sting) : Including target attack and non-target attack. + + Returns: + Tensor, adversarial image. + Tensor, mask image. + """ if attack_method == "non-target attack": - LOSS = FaceLossNoTargetAttack(self.target_emb) + loss = FaceLossNoTargetAttack(self.target_emb) if attack_method == "target_attack": - LOSS = FaceLossTargetAttack(self.target_emb) + loss = FaceLossTargetAttack(self.target_emb) - net_with_criterion = MyWithLossCell(self.resnet, LOSS, self.input_tensor) + net_with_criterion = MyWithLossCell(self.resnet, loss, self.input_tensor) train_net = MyTrainOneStepCell(net_with_criterion, self.opt) for i in range(2000): @@ -136,13 +143,13 @@ class FaceAdversarialAttack(object): def _reverse_norm(self, image_tensor): """ - Reverses normalization for a given image_tensor + Reverses normalization for a given image_tensor. Args: - image_tensor : Tensor + image_tensor (Tensor): image. Returns: - Tensor + Tensor, image. """ tensor = image_tensor * self.STD[:, None, None] + self.MEAN[:, None, None] @@ -157,12 +164,12 @@ class FaceAdversarialAttack(object): Apply a mask over an image. Args: - image_tensor : Canvas to be used to apply mask on. - mask_tensor : Mask to apply over the image. - reference_tensor : Used to reference mask boundaries + image_tensor (Tensor): Canvas to be used to apply mask on. + mask_tensor (Tensor): Mask to apply over the image. + reference_tensor (Tensor): Used to reference mask boundaries Returns: - Tensor + Tensor, image. """ tensor = mindspore.numpy.where((reference_tensor == 0), image_tensor, mask_tensor) @@ -173,10 +180,10 @@ class FaceAdversarialAttack(object): Create mask image. Args: - face_image : image of a detected face. + face_image (PIL.Image): image of a detected face. Returns: - mask_tensor : A mask image. + mask_tensor : a mask image. """ mask = Image.new('RGB', face_image.size, color=(0, 0, 0)) @@ -211,10 +218,10 @@ class FaceAdversarialAttack(object): Reverses normalization for a given image_tensor. Args: - image_tensor : Tensor. + image_tensor (Tensor): Tensor. Returns: - Tensor. + Tensor, image. """ tensor = image_tensor * self.STD[:, None, None] + self.MEAN[:, None, None] return tensor @@ -227,11 +234,10 @@ def detect_face(image_loc): detector. Args: - image_loc : image file location. + image_loc (numpy.ndarray): image file location. Returns: face_image : Resized face image. - """ detector = dlib.get_frontal_face_detector() @@ -253,7 +259,7 @@ def load_data(path_to_data): and test images. Args: - path_to_data : Path to the given data. + path_to_data (String): Path to the given data. Returns: list : List of resized face images. diff --git a/examples/community/face_adversarial_attack/camera.py b/examples/community/face_adversarial_attack/camera.py deleted file mode 100644 index 5ec3fae..0000000 --- a/examples/community/face_adversarial_attack/camera.py +++ /dev/null @@ -1,34 +0,0 @@ -# Copyright 2022 Huawei Technologies Co., Ltd -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -# ============================================================================ -import cv2 - - -capture = cv2.VideoCapture(0) - - -for i in range(2): - while(True): - ret, frame = capture.read() - width, height = capture.get(3), capture.get(4) - cv2.imwrite('./opencv_photo/input/input'+str(i)+'.png', frame) - cv2.imshow('frame', frame) - if cv2.waitKey(1) == ord('q'): - break - - -print(width, height) - -capture.release() -cv2.destroyAllWindows() \ No newline at end of file diff --git a/examples/community/face_adversarial_attack/example_non-target_attack.py b/examples/community/face_adversarial_attack/example_non-target_attack.py index 42d70e2..4861c4b 100644 --- a/examples/community/face_adversarial_attack/example_non-target_attack.py +++ b/examples/community/face_adversarial_attack/example_non-target_attack.py @@ -16,7 +16,7 @@ import numpy as np import matplotlib.image as mp from mindspore import context -import AFR +import ad context.set_context(mode=context.GRAPH_MODE, device_target="GPU") @@ -25,8 +25,8 @@ context.set_context(mode=context.GRAPH_MODE, device_target="GPU") if __name__ == '__main__': - inputs = AFR.load_data('opencv_photo/input/') - targets = AFR.load_data('opencv_photo/target/') + inputs = AFR.load_data('photos/input/') + targets = AFR.load_data('photos/target/') adversarial = AFR.FaceAdversarialAttack(inputs[0], targets[0]) @@ -34,7 +34,7 @@ if __name__ == '__main__': adversarial_tensor, mask_tensor = adversarial.train(attack_method) - mp.imsave('./outputs/对抗图像.jpg', np.transpose(adversarial_tensor.asnumpy(), (1, 2, 0))) - mp.imsave('./outputs/口罩.jpg', np.transpose(mask_tensor.asnumpy(), (1, 2, 0))) + mp.imsave('./outputs/adversarial_example.jpg', np.transpose(adversarial_tensor.asnumpy(), (1, 2, 0))) + mp.imsave('./outputs/mask.jpg', np.transpose(mask_tensor.asnumpy(), (1, 2, 0))) adversarial.test() diff --git a/examples/community/face_adversarial_attack/example_target_attack.py b/examples/community/face_adversarial_attack/example_target_attack.py index 7e089dd..dc24c4b 100644 --- a/examples/community/face_adversarial_attack/example_target_attack.py +++ b/examples/community/face_adversarial_attack/example_target_attack.py @@ -25,8 +25,8 @@ context.set_context(mode=context.GRAPH_MODE, device_target="GPU") if __name__ == '__main__': - inputs = AFR.load_data('opencv_photo/input/') - targets = AFR.load_data('opencv_photo/target/') + inputs = AFR.load_data('photos/input/') + targets = AFR.load_data('photos/target/') adversarial = AFR.FaceAdversarialAttack(inputs[0], targets[0]) @@ -35,7 +35,7 @@ if __name__ == '__main__': adversarial_tensor, mask_tensor = adversarial.train(attack_method) - mp.imsave('./outputs/对抗图像.jpg', np.transpose(adversarial_tensor.asnumpy(), (1, 2, 0))) - mp.imsave('./outputs/口罩.jpg', np.transpose(mask_tensor.asnumpy(), (1, 2, 0))) + mp.imsave('./outputs/adversarial_example.jpg', np.transpose(adversarial_tensor.asnumpy(), (1, 2, 0))) + mp.imsave('./outputs/mask.jpg', np.transpose(mask_tensor.asnumpy(), (1, 2, 0))) adversarial.test() diff --git a/examples/community/face_adversarial_attack/loss_design.py b/examples/community/face_adversarial_attack/loss_design.py index 2e70955..3441849 100644 --- a/examples/community/face_adversarial_attack/loss_design.py +++ b/examples/community/face_adversarial_attack/loss_design.py @@ -105,8 +105,9 @@ class FaceLossTargetAttack(nn.Cell): class FaceLossNoTargetAttack(nn.Cell): """The loss function of the non-target attack""" + def __init__(self, target_emb): - """初始化""" + """Initialization""" super(FaceLossNoTargetAttack, self).__init__() self.uniformreal = ops.UniformReal(seed=2) self.sum = ops.ReduceSum(keep_dims=False) diff --git a/examples/community/face_adversarial_attack/opencv_photo/input/input1.jpg b/examples/community/face_adversarial_attack/photos/input/input1.jpg similarity index 100% rename from examples/community/face_adversarial_attack/opencv_photo/input/input1.jpg rename to examples/community/face_adversarial_attack/photos/input/input1.jpg diff --git a/examples/community/face_adversarial_attack/opencv_photo/target/target1.jpg b/examples/community/face_adversarial_attack/photos/target/target1.jpg similarity index 100% rename from examples/community/face_adversarial_attack/opencv_photo/target/target1.jpg rename to examples/community/face_adversarial_attack/photos/target/target1.jpg diff --git a/examples/community/face_adversarial_attack/test.py b/examples/community/face_adversarial_attack/test.py index 7bbc62b..de483d8 100644 --- a/examples/community/face_adversarial_attack/test.py +++ b/examples/community/face_adversarial_attack/test.py @@ -33,9 +33,9 @@ if __name__ == '__main__': The input image, target image and adversarial image are tested using the FaceRecognition model. """ - image = AFR.load_data('opencv_photo/adv_input') - inputs = AFR.load_data('opencv_photo/input/') - targets = AFR.load_data('opencv_photo/target/') + image = AFR.load_data('photos/adv_input/') + inputs = AFR.load_data('photos/input/') + targets = AFR.load_data('photos/target/') tensorize = ToTensor() normalize = P.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) @@ -45,7 +45,7 @@ if __name__ == '__main__': resnet = get_net() - image = mp.imread("./对抗图像.jpg") + image = mp.imread("./outputs/adversarial_example.jpg") adv = Tensor(normalize(tensorize(image))) input_tensor = Tensor(normalize(tensorize(inputs[0])))