Return the poisoned dataset.

Test the victim model. Args: test_dl1 (torch.utils.data.DataLoader): Benign testing dataloader test_dl2 (torc

Return the victim model.

r""" Save adversarial images as torch.tensor from given torch.utils.data.DataLoader. Arguments: save_path (str) : save_pat

Perform NAD defense method based on attacked models. The repaired model will be stored in self.model Args: datas

Perform ABL defense method based on attacked models. The repaired model will be stored in self.model Args: split

Computes the precision@k for the specified values of k

Uniform test API for any model and any dataset. Args: model (torch.nn.Module): Network. dataset (torch.utils.data.Dataset): Datas

for image size 32, feature_dim = 512 for other sizes, feature_dim = 512 * (size//32)**2

test BA and ASR after MCR. Args: model_name (str): name of the network dataset_name (str): name of the dataset attack

test BA and ASR before MCR. Args: start_model (nn.Module): Start model of MCR, params needs to be loaded beforehand. end_model (n

This class is used to generating UAP given a benign dataset and a benign model. :param model: Benign model. :param t

Add reflection-based trigger to images. Args: sample (torch.Tensor): shape (C,H,W), index (interger): index o

Each image in dataset should be torch.Tensor, shape (C,H,W)

Each image in dataset should be torch.Tensor, shape (C,H,W)

Test curve model on test dataset Args: dataset (types in support_list): Dataset. schedule (dict): Schedule for testin

train model using given schedule and test with given datasets

Perform unet defense method on data and return the preprocessed data. Args: data (torch.Tensor): Input data (between 0.0 and 1.0)

Analytical solution for min_{c} |cx_1 + (1-c)x_2|_2^2 d is the distance (objective) optimzed v1v1 = <x1,x1> v1v2 = <x

The basic training function,

Computes the precision@k for the specified values of k

Add watermarked trigger to image. Args: img (torch.Tensor): shape (C, H, W). Returns: torch.Tensor: Poisoned

Generate an identity grid with shape 1*height*height*2 and a noise grid with shape 1*height*height*2 according to the input height ``height`` and

Add patch to images, and change label to target label, set random_path to True if patch in random localtion

Convert a ``PIL Image`` to a tensor of the same type. This function does not support torchscript. See :class:`~torchvision.transforms.PILToTe

for image size 32, feature_dim = 512 for other sizes, feature_dim = 512 * (size//32)**2

Convert a strpath, PIL.Image.Image, numpy.ndarray, torch.Tensor object to a torch.Tensor object. Args: x (strpath | PIL.Image.Image | num

A generic function to create a poisoned dataset using any AttributeEditor.

Find the minimum norm solution as combination of two points This is correct only in 2D ie. min_c |\sum c_i x_i|_2^2 st. \sum

A utility function to create a file-safe string from text.

r""" Function for changing the training mode of the model.

Add watermarked trigger to image. Args: img (torch.Tensor): shape (C, H, W). Returns: torch.Tensor: Poisoned

Add watermarked trigger to image. Args: img (torch.Tensor): shape (C, H, W). Returns: torch.Tensor: Poisoned

Add WaNet trigger to image. Args: img (torch.Tensor): shape (C, H, W). noise (bool): turn on noise mode, default is F

Add watermarked trigger to image. Args: img (torch.Tensor): shape (C, H, W). Returns: torch.Tensor: Poisoned

Add watermarked trigger to image. Args: img (torch.Tensor): shape (C, H, W). Returns: torch.Tensor: Poisoned