Manual convolution with same padding Although PyTorch >= 1.10.0 supports ``padding='same'`` as a keyword argument, this does not export to Co

Downscaling with maxpool then double conv

Upscaling then double conv

(convolution => [BN] => ReLU) * 2

Construct a dataset. Args: data_list (list): the list of data. labels (list): the list of label.

The Conv2dBlock in the generator of dynamic backdoor trigger.

Construct a dataset. Args: data_list (list): the list of data. labels (list): the list of label.

Construct a dataset. Args: data_list (list): the list of data. labels (list): the list of label.

Construct a dataset. Args: data_list (list): the list of data. labels (list): the list of label.

Normalization of images. Args: dataset_name (str): the name of the dataset to be normalized. expected_values (float): the normali

Add watermarked trigger to CIFAR10 image. Args: pattern (None | torch.Tensor): shape (3, 32, 32) or (32, 32). weight (None | torc

Add WaNet trigger to CIFAR10 image. Args: identity_grid (orch.Tensor): the poisoned pattern shape. noise_grid (orch.Tensor): the

Add watermarked trigger to DatasetFolder images. Args: pattern (torch.Tensor): shape (C, H, W) or (H, W). weight (torch.Tensor):

Add watermarked trigger to DatasetFolder images. Args: pattern (torch.Tensor): shape (C, H, W) or (H, W). weight (torch.Tensor):

Add WaNet trigger to DatasetFolder images. Args: identity_grid (orch.Tensor): the poisoned pattern shape. noise_grid (orch.Tensor

Add watermarked trigger to MNIST image. Args: pattern (None | torch.Tensor): shape (1, 28, 28) or (28, 28). weight (None | torch.

Add WaNet trigger to MNIST image. Args: identity_grid (orch.Tensor): the poisoned pattern shape. noise_grid (orch.Tensor): the no

The generator of backdoor trigger on GTSRB.

Denormalization of images. Args: dataset_name (str): the name of the dataset to be denormalized. expected_values (float): the den

The generator of dynamic backdoor trigger. Args: dataset_name (str): the name of the dataset. out_channels (int): the output

The generator of backdoor trigger on MNIST.

Normalization of images. Args: dataset_name (str): the name of the dataset to be normalized. expected_values (float): the normali

The data augmentation transform by the probability. Args: f (nn.Module): the data augmentation transform operation. p (float): th

The data augmentation transform by the probability. Args: f (nn.Module): the data augmentation transform operation. p (float)

The image steganography decoder to assist the training of the image steganography encoder. We implement it based on the official tensorflow versi

The image steganography encoder to implant the backdoor trigger. We implement it based on the official tensorflow version: https://github.co

Supervised Contrastive Learning: https://arxiv.org/pdf/2004.11362.pdf. It also supports the unsupervised contrastive loss in SimCLR

A simple loading dataset - loads the tensor that are passed in input. This is the same as torch.utils.data.TensorDataset except that you can

The generator of backdoor trigger on CIFAR10.

Paying More Attention to Attention: Improving the Performance of Convolutional Neural Netkworks via Attention Transfer https://arxiv.org/pdf/1612.

Add watermarked trigger to CIFAR10 image. Args: pattern (None | torch.Tensor): shape (3, 32, 32) or (32, 32). weight (None | torc

Add watermarked trigger to CIFAR10 image. Args: pattern (None | torch.Tensor): shape (3, 32, 32) or (32, 32). weight (None | torc

Add watermarked trigger to CIFAR10 image. Args: pattern (None | torch.Tensor): shape (3, 32, 32) or (32, 32). weight (None | torc

Add watermarked trigger to CIFAR10 image. Args: pattern (None | torch.Tensor): shape (3, 32, 32) or (32, 32). weight (None | torc

Add watermarked trigger to DatasetFolder images. Args: pattern (torch.Tensor): shape (C, H, W) or (H, W). weight (torch.Tensor):

Add watermarked trigger to DatasetFolder images. Args: pattern (torch.Tensor): shape (C, H, W) or (H, W). weight (torch.Tensor):

Add watermarked trigger to DatasetFolder images. Args: pattern (torch.Tensor): shape (C, H, W) or (H, W). weight (torch.Tensor):

Add watermarked trigger to MNIST image. Args: pattern (None | torch.Tensor): shape (1, 28, 28) or (28, 28). weight (None | torch.

Add watermarked trigger to MNIST image. Args: pattern (None | torch.Tensor): shape (1, 28, 28) or (28, 28). weight (None | torch.

Add watermarked trigger to MNIST image. Args: pattern (None | torch.Tensor): shape (1, 28, 28) or (28, 28). weight (None | torch.

Add watermarked trigger to MNIST image. Args: pattern (None | torch.Tensor): shape (1, 28, 28) or (28, 28). weight (None | torch.

Autoencoder for 1x28x28 input image. This is a reimplementation of the blog post 'Building Autoencoders in Keras', from blog `Building Autoen

Autoencoder for 3x32x32 input image. This is modified from 'PyTorch-CIFAR-10-autoencoder', from github `PyTorch-CIFAR-10-autoencoder <https:/

Baseline network for MNIST dataset. This network is the implement of baseline network for MNIST dataset, from paper `BadNets: Evaluating Back

Randomly mask out one or more patches from an image. Args: n_holes (int): Number of patches to cut out of each image. length (int)

Dataset that poison original dataset by adding small perturbation (delta) to original dataset, and changing label to target label (t_lable) Thi

The image steganography discriminator to assist the training of the image steganography encoder and decoder. We implement it based on the officia

The DownSampleBlock in the generator of dynamic backdoor trigger.

The image steganography discriminator to assist the training of the image steganography encoder and decoder (Customized for MNIST dataset). We im

The image steganography decoder to assist the training of the image steganography encoder (Customized for MNIST dataset). We implement it based o

The image steganography encoder to implant the backdoor trigger (Customized for MNIST dataset). We implement it based on the official tensorflow

r""" PGD in the paper 'Towards Deep Learning Models Resistant to Adversarial Attacks' [https://arxiv.org/abs/1706.06083] Distance Mea