Edit a single image based on the provided attribute description and save it. If `attribute_description` is None, the pre-computed lat

Test the backdoor trigger mask generator. Args: modelM (torch.nn.Module): Backdoor trigger mask generator. te

Extend source_set to #source_num samples, allowing more samples to poison

filter out poisoned samples from poisoned dataset. Args: schedule (dict): schedule for spliting the dataset.

Given a list of vectors (vecs), this method finds the minimum norm element in the convex hull as min |u|_2 st. u = \sum c_i vecs[i]

Test the victim model using the backdoor trigger generator in training process. Args: atkmodel (torch.nn.Module): Backdoo

r""" It defines the computation performed at every call. Should be overridden by all subclasses.

Frozen the layers which don't need to fine tuning.

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

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

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

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

Get the encoded images with the trigger pattern. Args: path (str): The path of the saved image steganography encoder.

Return the backdoor trigger pattern generator.

Return the backdoor trigger mask generator.

Compute the blind passenger loss term.

uniformly initialize perturbation that will add to selected target image

Add poison_deltas to poisoned_samples

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

Perform ShrinkPad defense method on data and return the preprocessed data. Args: data (torch.Tensor): Input data. siz

Processes a directory of images, applying the attribute edit to a subset of them. This is the "Template Method" that uses the abstrac

Project on the lp ball centered at 0 and of radius epsilon, SUPPORTS only p = 2 and p = Inf for now :param perturbation: Pert

https://discuss.pytorch.org/t/cropping-a-minibatch-of-images-each-image-a-bit-differently/12247/5.

pruning. Args: schedule (dict): Schedule for testing.

Finetuning. Args: schedule (dict): Schedule for testing.

select samples from target class with large gradients

r""" Set the return type of adversarial images: `int` or `float`. Arguments: type (str) : 'float' or 'int'. (DEFAULT : 'fl

Split dataset into poisoned part and clean part. The ratio of poisoned part is controlled by split_ratio. Args: dataset (

first pretrain/load a mdoel, then use to craft poisoned dataset under the priciple of gradient alignment, then used the poisoned dataset to train a ne

Train the backdoor trigger mask generator. Args: modelM (torch.nn.Module): Backdoor trigger mask generator. o

Train the victim model and the backdoor trigger generator jointly. Args: atkmodel (torch.nn.Module): Backdoor trigger gen

Train the victim model and the backdoor trigger pattern generator jointly. Args: model (torch.nn.Module): Victim model.

:param delta: controls the desired fooling rate (default = 80% fooling rate) :param max_iter_uni: optional other termination criterio

Get the poisoned image. Args: img (PIL.Image.Image | numpy.ndarray | torch.Tensor): If img is numpy.ndarray or torch.Tensor, the

Get the poisoned image. Args: img (PIL.Image.Image | numpy.ndarray | torch.Tensor): If img is numpy.ndarray or torch.Tensor, the

Get the poisoned image. Args: img (PIL.Image.Image | numpy.ndarray | torch.Tensor): If img is numpy.ndarray or torch.Tensor, the

Get the poisoned image. Args: img (PIL.Image.Image | numpy.ndarray | torch.Tensor): If img is numpy.ndarray or torch.Tensor, the

Get the poisoned image. Args: img (PIL.Image.Image | numpy.ndarray | torch.Tensor): If img is numpy.ndarray or torch.Tensor, the

Get the poisoned image. Args: img (PIL.Image.Image | numpy.ndarray | torch.Tensor): If img is numpy.ndarray or torch.Tensor, the