Data instance which defines the basic attributes. Args: impath (str): image path. label (int): class label. domain (int):

A simple neural network composed of a CNN backbone and optionally a head such as mlp for classification.

A registry providing name -> object mapping, to support custom modules. To create a registry (e.g. a backbone registry): .. code-block::

STL-10 dataset. Description: - 10 classes: airplane, bird, car, cat, deer, dog, horse, monkey, ship, truck. - Images are 96x96 pixels

Compute and store the average and current value. Examples:: >>> # 1. Initialize a meter to record loss >>> losses = AverageMeter(

CIFAR10 for SSL. Reference: - Krizhevsky. Learning Multiple Layers of Features from Tiny Images. Tech report.

Fully convolutional network.

Gradient reversal layer. It acts as an identity layer in the forward, but reverses the sign of the gradient in the backward.

CIFAR100 for SSL. Reference: - Krizhevsky. Learning Multiple Layers of Features from Tiny Images. Tech report.

Normalize data using per-channel mean and standard deviation. Reference: - Ulyanov et al. Instance normalization: The missing in- gredien

Mobile Inverted Residual Bottleneck Block Args: block_args (namedtuple): BlockArgs, see above global_params (namedtuple): Gl

Store the average and current value for a set of metrics. Examples:: >>> # 1. Create an instance of MetricMeter >>> metric = Metr

SVHN for SSL. Reference: - Netzer et al. Reading Digits in Natural Images with Unsupervised Feature Learning. NIPS-W 2011.

Attention from `"Dynamic Domain Generalization" <https://github.com/MetaVisionLab/DDG>`_.

Randomly choose one of the best 25 Sub-policies on CIFAR10. Example: >>> policy = CIFAR10Policy() >>> transformed = policy(image)

Conv2dDynamic from `"Dynamic Domain Generalization" <https://github.com/MetaVisionLab/DDG>`_.

Randomly mask out one or more patches from an image. https://github.com/uoguelph-mlrg/Cutout Args: n_holes (int, optional): number o

Add gaussian noise.

Randomly choose one of the best 24 Sub-policies on ImageNet. Example: >>> policy = ImageNetPolicy() >>> transformed = policy(imag

Localization network.

Write console output to external text file. Imported from `<https://github.com/Cysu/open-reid/blob/master/reid/utils/logging.py>`_ Args:

Given an image of (height, width), we resize it to (height*1.125, width*1.125), and then perform random cropping. Args: height (int):

Randomly samples N classes each with K instances to form a minibatch of size N*K. Modified from https://github.com/KaiyangZhou/deep-person-re

Randomly samples N domains each with K images to form a minibatch of size N*K. Args: data_source (list): list of Datums. batc

Randomly choose one of the best 25 Sub-policies on SVHN. Example: >>> policy = SVHNPolicy() >>> transformed = policy(image)

Sequential domain sampler, which randomly samples K images from each domain to form a minibatch. Args: data_source (list): list of Da

Adversarial Discriminative Domain Adaptation. https://arxiv.org/abs/1702.05464.

Adaptive Batch Normalization. https://arxiv.org/abs/1603.04779.

Block Decoder for readability, straight from the official TensorFlow repository

Cross Domain Adaptive Clustering. https://arxiv.org/pdf/2104.09415.pdf

CIFAR-100 -> CIFAR-100-C. Dataset link: https://zenodo.org/record/3555552#.YFxpQmQzb0o Statistics: - 2 domains: the normal CIFAR-100

CIFAR-10 -> CIFAR-10-C. Dataset link: https://zenodo.org/record/2535967#.YFwtV2Qzb0o Statistics: - 2 domains: the normal CIFAR-10 vs

CIFAR-10 and STL-10. CIFAR-10: - 60,000 32x32 colour images. - 10 classes, with 6,000 images per class. - 50,000 training

Tumor tissue recognition. 2 classes (whether a given region of tissue contains tumor tissue). Reference: - Bandi et al. "From detect

Evaluator for classification.

2D Convolutions like TensorFlow, for a dynamic image size

2D Convolutions like TensorFlow, for a fixed image size

Cross-gradient training. https://arxiv.org/abs/1804.10745.

Reference : https://github.com/quark0/darts/blob/master/cnn/utils.py

Domain Adaptive Ensemble Learning. https://arxiv.org/abs/2003.07325.

Domain Adaptive Ensemble Learning. DG version: only use labeled source data. https://arxiv.org/abs/2003.07325.

Domain-Adversarial Neural Networks. https://arxiv.org/abs/1505.07818.

Deep Domain-Adversarial Image Generation. https://arxiv.org/abs/2003.06054.

A unified dataset class for 1) domain adaptation 2) domain generalization 3) semi-supervised learning

Five digit datasets. It contains: - MNIST: hand-written digits. - MNIST-M: variant of MNIST with blended background. - SV

Digit recognition datasets for single-source domain generalization. There are five digit datasets: - MNIST: hand-written digits.

Digits-DG. It contains 4 digit datasets: - MNIST: hand-written digits. - MNIST-M: variant of MNIST with blended background.

DomainMix. Dynamic Domain Generalization. https://github.com/MetaVisionLab/DDG

DomainNet. Statistics: - 6 distinct domains: Clipart, Infograph, Painting, Quickdraw, Real, Sketch. - Around 0.6M images.

EFDMix. Reference: Zhang et al. Exact Feature Distribution Matching for Arbitrary Style Transfer and Domain Generalization. CVPR 2022.

An EfficientNet model. Most easily loaded with the .from_name or .from_pretrained methods Args: blocks_args (list): A list of BlockA

Entropy Minimization. http://papers.nips.cc/paper/2740-semi-supervised-learning-by-entropy-minimization.pdf.

Base evaluator.

Satellite imagery classification. 62 classes (building or land use categories). Reference: - Christie et al. "Functional Map of the

FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence. https://arxiv.org/abs/2001.07685.

Animal species recognition. 182 classes (species). Reference: - Beery et al. "The iwildcam 2021 competition dataset." arXiv 2021.