normal

class trojanvision.datasets.MNIST(norm_par={'mean': [0.1307], 'std': [0.3081]}, **kwargs)

MNIST dataset. It inherits trojanvision.datasets.ImageSet.

See also

• torchvision: torchvision.datasets.MNIST

• paper: Gradient-based learning applied to document recognition

• website: http://yann.lecun.com/exdb/mnist/

Variables:

• name (str) – 'mnist'

• num_classes (int) – 10

• data_shape (list[int]) – [1, 28, 28]

• norm_par (dict[str, list[float]]) – {'mean': [0.1307], 'std': [0.3081]}

class trojanvision.datasets.CIFAR10(norm_par={'mean': [0.49139968, 0.48215827, 0.44653124], 'std': [0.24703233, 0.24348505, 0.26158768]}, **kwargs)

CIFAR10 dataset introduced by Alex Krizhevsky in 2009. It inherits trojanvision.datasets.ImageSet.

See also

• torchvision: torchvision.datasets.CIFAR10

• paper: Learning Multiple Layers of Features from Tiny Images

• website: https://www.cs.toronto.edu/~kriz/cifar.html

Variables:

• name (str) – 'cifar10'

• num_classes (int) – 10

• data_shape (list[int]) – [3, 32, 32]

• class_names (list[str]) –

  ['airplane', 'automobile', 'bird', 'cat', 'deer',

  'dog', 'frog', 'horse', 'ship', 'truck']

• norm_par (dict[str, list[float]]) –

  {'mean': [0.49139968, 0.48215827, 0.44653124],

  'std'  : [0.24703233, 0.24348505, 0.26158768]}

class trojanvision.datasets.CIFAR100(norm_par={'mean': [0.49139968, 0.48215827, 0.44653124], 'std': [0.24703233, 0.24348505, 0.26158768]}, **kwargs)

CIFAR100 dataset. It inherits trojanvision.datasets.ImageSet.

See also

• torchvision: torchvision.datasets.CIFAR100

• paper: Learning Multiple Layers of Features from Tiny Images

• website: https://www.cs.toronto.edu/~kriz/cifar.html

Variables:

• name (str) – 'cifar100'

• num_classes (int) – 100

• data_shape (list[int]) – [3, 32, 32]

• norm_par (dict[str, list[float]]) –

  {'mean': [0.49139968, 0.48215827, 0.44653124],

  'std'  : [0.24703233, 0.24348505, 0.26158768]}

class trojanvision.datasets.ImageNet16(norm_par={'mean': [122.68 / 255, 116.66 / 255, 104.01 / 255], 'std': [63.22 / 255, 61.26 / 255, 65.09 / 255]}, num_classes=1000, **kwargs)

ImageNet16 dataset introduced by Patryk Chrabaszcz in 2017. It inherits trojanvision.datasets.ImageSet.

See also

• paper: A Downsampled Variant of ImageNet as an Alternative to the CIFAR datasets

• website: https://patrykchrabaszcz.github.io/Imagenet32/

Variables:

• name (str) – 'imagenet16'

• num_classes (int) – Flexible (passed by command line argument, no larger than 1000).

• data_shape (list[int]) – [3, 16, 16]

class trojanvision.datasets.ImageNet32(norm_par={'mean': [122.68 / 255, 116.66 / 255, 104.01 / 255], 'std': [63.22 / 255, 61.26 / 255, 65.09 / 255]}, num_classes=1000, **kwargs)

ImageNet32 dataset introduced by Patryk Chrabaszcz in 2017. It inherits trojanvision.datasets.ImageSet.

See also

• paper: A Downsampled Variant of ImageNet as an Alternative to the CIFAR datasets

• website: https://patrykchrabaszcz.github.io/Imagenet32/

Variables:

• name (str) – 'imagenet64'

• num_classes (int) – Flexible (passed by command line argument, no larger than 1000).

• data_shape (list[int]) – [3, 32, 32]

class trojanvision.datasets.STL10(norm_par={'mean': [0.507, 0.487, 0.441], 'std': [0.267, 0.256, 0.276]}, **kwargs)

STL10 dataset. It inherits trojanvision.datasets.ImageSet.

See also

• torchvision: torchvision.datasets.STL10

• paper: An Analysis of Single-Layer Networks in Unsupervised Feature Learning

• website: https://cs.stanford.edu/~acoates/stl10/

Variables:

• name (str) – 'stl10'

• num_classes (int) – 10

• data_shape (list[int]) – [3, 256, 256]

• norm_par (dict[str, list[float]]) – {'mean': [0.507, 0.487, 0.441], 'std': [0.267, 0.256, 0.276]}

get_transform(mode, normalize=None)

Get dataset transform.

Parameters:

• mode (str) – The dataset mode (e.g., 'train' | 'valid' | 'unlabeled' | 'train+unlabeled').

• normalize (bool | None) – Whether to use torchvision.transforms.Normalize in dataset transform. Defaults to self.normalize.

Returns:

torchvision.transforms.Compose – The transform sequence.