Awesome Domain Generalization

This repository is a collection of awesome things about domain generalization, including papers, code, etc.

If you would like to contribute to our repository or have any questions/advice, see Contributing & Contact.

Awesome Domain Generalization
Contents
Papers
Publications
Datasets
Libraries
Lectures & Tutorials & Talks
Other Resources
Contributing & Contact
Acknowledgements

Papers

We list papers, implementation code (the unofficial code is marked with *), etc, in the order of year and from journals to conferences. Note that some papers may fall into multiple categories.

Survey

Generalizing to Unseen Domains: A Survey on Domain Generalization [IJCAI 2021] [Slides] [155]
Domain Generalization in Vision: A Survey [TPAMI 2022] [3]

Theory & Analysis

We list the papers that either provide inspiring theoretical analyses or conduct extensive empirical studies for domain generalization.

A Generalization Error Bound for Multi-Class Domain Generalization [arXiv 2019] [123]
Domain Generalization by Marginal Transfer Learning [JMLR 2021] [Code] (MTL) [188]
The Risks of Invariant Risk Minimization [ICLR 2021] [196]
In Search of Lost Domain Generalization [ICLR 2021] [134]
The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization [ICCV 2021] [Code] [135]
An Empirical Investigation of Domain Generalization with Empirical Risk Minimizers [NeurIPS 2021] [Code] [198]
Towards a Theoretical Framework of Out-Of-Distribution Generalization [NeurIPS 2021] [199]
Out-of-Distribution Generalization in Kernel Regression [NeurIPS 2021] [205]
Quantifying and Improving Transferability in Domain Generalization [NeurIPS 2021] [Code] (Transfer) [206]
OoD-Bench: Quantifying and Understanding Two Dimensions of Out-of-Distribution Generalization [CVPR 2022] [Code] (OoD-Bench) [214]

Dataset

Free Viewpoint Action Recognition Using Motion History Volumes [CVIU 2006] (IXMAS dataset) [39]
Geodesic flow kernel for unsupervised domain adaptation [CVPR 2012] (Office-Caltech dataset) [32]
Unbiased Metric Learning: On the Utilization of Multiple Datasets and Web Images for Softening Bias [ICCV 2013] (VLCS dataset) [16]
Domain Generalization for Object Recognition with Multi-Task Autoencoders [ICCV 2015] [Code] (MTAE, Rotated MNIST dataset) [6]
Scalable Person Re-identification: A Benchmark [ICCV 2015] (Market-1501 dataset) [46]
The Cityscapes Dataset for Semantic Urban Scene Understanding [CVPR 2016] (Cityscapes dataset) [44]
The SYNTHIA Dataset: A Large Collection of Synthetic Images for Semantic Segmentation of Urban Scenes [CVPR 2016] (SYNTHIA dataset) [42]
Playing for Data: Ground Truth from Computer Games [ECCV 2016] (GTA5 dataset) [43]
Performance Measures and a Data Set for Multi-target, Multi-camera Tracking [ECCV 2016] (Duke dataset) [47]
VisDA: The Visual Domain Adaptation Challenge [arXiv 2017] (VisDA-17 dataset) [36]
Deep Hashing Network for Unsupervised Domain Adaptation [CVPR 2017] (OfficeHome dataset) [20]
Deeper, Broader and Artier Domain Generalization [ICCV 2017] [Code] (PACS dataset) [2]
Learning Multiple Visual Domains with Residual Adapters [NeurIPS 2017] (Visual Decathlon (VD) dataset) [38]
Recognition in Terra Incognita [ECCV 2018] (Terra Incognita dataset) [45]
Invariant Risk Minimization [arXiv 2019] [Code] (IRM, Colored MNIST dataset) [165]
Learning Robust Representations by Projecting Superficial Statistics Out [ICLR 2019] [Code] (HEX, ImageNet-Sketch dataset) [35]
Benchmarking Neural Network Robustness to Common Corruptions and Perturbations [ICLR 2019] (CIFAR-10-C / CIFAR-100-C / ImageNet-C dataset) [37]
Moment Matching for Multi-Source Domain Adaptation [ICCV 2019] [Code] (DomainNet dataset) [33]
Learning to Generate Novel Domains for Domain Generalization [ECCV 2020] [Code] (L2A-OT, Digits-DG dataset) [28]
Domain Adaptive Ensemble Learning [TIP 2021] [Code] (mini-DomainNet dataset) [34]
Towards Non-IID Image Classification A Dataset and Baselines [PR 2021] (NICO dataset) [108]
NICO++ Towards Better Benchmarking for Domain Generalization [arXiv 2022] (NICO++ dataset) [183]
MetaShift: A Dataset of Datasets for Evaluating Contextual Distribution Shifts and Training Conflicts [ICLR 2022] [Code] (MetaShift dataset) [213]

Domain Generalization

To address the dataset/domain shift problem [109]) [110)] [111)] [112], domain generalization [113] aims to learn a model from source domain(s) and make it generalize well to unknown target domains.

Domain Alignment-Based Methods

Domain alignment-based methods aim to minimize divergence between source domains for learning domain-invariant representations.

Domain Generalization via Invariant Feature Representation [ICML 2013] [Code] (DICA) [65]
Domain-Adversarial Training of Neural Networks [JMLR 2016] [Code] (DANN) [226]
Learning Attributes Equals Multi-Source Domain Generalization [CVPR 2016] (UDICA) [120]
Robust Domain Generalisation by Enforcing Distribution Invariance [IJCAI 2016] (ESRand) [66]
Scatter Component Analysis A Unified Framework for Domain Adaptation and Domain Generalization [TPAMI 2017] (SCA) [67]
Unified Deep Supervised Domain Adaptation and Generalization [ICCV 2017] [Code] (CCSA) [71]
Beyond Domain Adaptation: Unseen Domain Encapsulation via Universal Non-volume Preserving Models [arXiv 2018] (UNVP) [166]
Domain Generalization via Conditional Invariant Representation [AAAI 2018] (CIDG) [68]
Domain Generalization with Adversarial Feature Learning [CVPR 2018] [Code] (MMD-AAE) [76]
Deep Domain Generalization via Conditional Invariant Adversarial Networks [ECCV 2018] (CIDDG, CDANN) [77]
Generalizing to Unseen Domains via Distribution Matching [arXiv 2019] [Code] (G2DM) [81]
Image Alignment in Unseen Domains via Domain Deep Generalization [arXiv 2019] (DeGIA) [169]
Multi-Adversarial Discriminative Deep Domain Generalization for Face Presentation Attack Detection [CVPR 2019] [Code] (MADDG) [78]
Generalizable Feature Learning in the Presence of Data Bias and Domain Class Imbalance with Application to Skin Lesion Classification [MICCAI 2019] [72]
Domain Generalization via Model-Agnostic Learning of Semantic Features [NeurIPS 2019] [Code] (MASF) [18]
Adversarial Invariant Feature Learning with Accuracy Constraint for Domain Generalization [ECMLPKDD 2019] [Code] (AFLAC) [84]
Feature Alignment and Restoration for Domain Generalization and Adaptation [arXiv 2020] (FAR) [189]
Representation via Representations: Domain Generalization via Adversarially Learned Invariant Representations [arXiv 2020] (RVR) [82]
Correlation-aware Adversarial Domain Adaptation and Generalization [PR 2020] [Code] (CAADA) [80]
Domain Generalization Using a Mixture of Multiple Latent Domains [AAAI 2020] [Code] [83]
Single-Side Domain Generalization for Face Anti-Spoofing [CVPR 2020] [Code] (SSDG) [79]
Scanner Invariant Multiple Sclerosis Lesion Segmentation from MRI [ISBI 2020] [85]
Respecting Domain Relations: Hypothesis Invariance for Domain Generalization [ICPR 2020] (HIR) [74]
Domain Generalization via Multidomain Discriminant Analysis [UAI 2020] [Code] (MDA) [70]
Domain Generalization for Medical Imaging Classification with Linear-Dependency Regularization [NeurIPS 2020] [Code] (LDDG) [75]
Domain Generalization via Entropy Regularization [NeurIPS 2020] [Code] [86]
Iterative Feature Matching: Toward Provable Domain Generalization with Logarithmic Environments [arXiv 2021] [192]
Semi-Supervised Domain Generalization in RealWorld: New Benchmark and Strong Baseline [arXiv 2021] [179]
Collaborative Semantic Aggregation and Calibration for Separated Domain Generalization [arXiv 2021] [Code] (CSAC) [161]
Multi-Domain Adversarial Feature Generalization for Person Re-Identification [TIP 2021] (MMFA-AAE) [144]
Scale Invariant Domain Generalization Image Recapture Detection [ICONIP 2021] (SADG) [177]
Domain Generalization under Conditional and Label Shifts via Variational Bayesian Inference [IJCAI 2021] (VBCLS) [195]
Domain Generalization using Causal Matching [ICML 2021] [Code] (MatchDG) [73]
Generalization on Unseen Domains via Inference-Time Label-Preserving Target Projections [CVPR 2021] [Code] [118]
Progressive Domain Expansion Network for Single Domain Generalization [CVPR 2021] [Code] (PDEN) [141]
Confidence Calibration for Domain Generalization Under Covariate Shift [ICCV 2021] [133]
On Calibration and Out-of-domain Generalization [NeurIPS 2021] [154]
Domain-invariant Feature Exploration for Domain Generalization [TMLR 2022] [Code] (DIFEX) [209]

Data Augmentation-Based Methods

Data augmentation-based methods augment original data and train the model on the generated data to improve model robustness.

Certifying Some Distributional Robustness with Principled Adversarial Training [arXiv 2017] [Code] [52]
Generalizing across Domains via Cross-Gradient Training [ICLR 2018] [Code] (CrossGrad) [53]
Generalizing to Unseen Domains via Adversarial Data Augmentation [NeurIPS 2018] [Code] [25]
Staining Invariant Features for Improving Generalization of Deep Convolutional Neural Networks in Computational Pathology [Frontiers in Bioengineering and Biotechnology 2019] [26]
Multi-component Image Translation for Deep Domain Generalization [WACV 2019] [Code] [167]
Domain Generalization by Solving Jigsaw Puzzles [CVPR 2019] [Code] (JiGen) [98]
Addressing Model Vulnerability to Distributional Shifts Over Image Transformation Sets [ICCV 2019] [Code] [21]
Domain Randomization and Pyramid Consistency: Simulation-to-Real Generalization Without Accessing Target Domain Data [ICCV 2019] [Code] [62]
Hallucinating Agnostic Images to Generalize Across Domains [ICCV workshop 2019] [Code] [63]
Improve Unsupervised Domain Adaptation with Mixup Training [arXiv 2020] [Code*] (Mixup) [227]
Improving the Generalizability of Convolutional Neural Network-Based Segmentation on CMR Images [Frontiers in Cardiovascular Medicine 2020] [24]
Generalizing Deep Learning for Medical Image Segmentation to Unseen Domains via Deep Stacked Transformation [TMI 2020] (BigAug) [23]
Deep Domain-Adversarial Image Generation for Domain Generalisation [AAAI 2020] [Code] (DDAIG) [55]
Towards Universal Representation Learning for Deep Face Recognition [CVPR 2020] [Code] [22]
Heterogeneous Domain Generalization via Domain Mixup [ICASSP 2020] [Code] [128]
Learning to Generate Novel Domains for Domain Generalization [ECCV 2020] [Code] (L2A-OT, Digits-DG dataset) [28]
Learning from Extrinsic and Intrinsic Supervisions for Domain Generalization [ECCV 2020] [Code] (EISNet) [99]
Towards Recognizing Unseen Categories in Unseen Domains [ECCV 2020] [Code] (CuMix) [57]
Rethinking Domain Generalization Baselines [ICPR 2020]
More is Better: A Novel Multi-view Framework for Domain Generalization [arXiv 2021] [184]
Semi-Supervised Domain Generalization with Stochastic StyleMatch [arXiv 2021] [Code] (StyleMatch) [54]
Better Pseudo-label Joint Domain-aware Label and Dual-classifier for Semi-supervised Domain Generalization [arXiv 2021] [156]
Out-of-domain Generalization from a Single Source: A Uncertainty Quantification Approach [arXiv 2021] [151]
Towards Principled Disentanglement for Domain Generalization [arXiv 2021] [Code] (DDG) [170]
MixStyle Neural Networks for Domain Generalization and Adaptation [arXiv 2021] [Code] (MixStyle) [58]
VideoDG: Generalizing Temporal Relations in Videos to Novel Domains [TPAMI 2021] [Code] (APN) [197]
Domain Generalization by Marginal Transfer Learning [JMLR 2021] [Code] [188]
Domain Generalisation with Domain Augmented Supervised Contrastive Learning [AAAI Student Abstract 2021] (DASCL) [139]
DecAug: Out-of-Distribution Generalization via Decomposed Feature Representation and Semantic Augmentation [AAAI 2021] [Code] (DecAug) [171]
Domain Generalization with Mixstyle [ICLR 2021] [Code] (MixStyle) [56]
Robust and Generalizable Visual Representation Learning via Random Convolutions [ICLR 2021] [Code] (RC) [59]
Learning to Learn Single Domain Generalization [CVPR 2020] [Code] (M-ADA) [27]
FSDR: Frequency Space Domain Randomization for Domain Generalization [CVPR 2021] [Code] (FSDR) [115]
FedDG: Federated Domain Generalization on Medical Image Segmentation via Episodic Learning in Continuous Frequency Space [CVPR 2021] [Code] (FedDG) [147]
Uncertainty-guided Model Generalization to Unseen Domains [CVPR 2021] [Code] [168]
Continual Adaptation of Visual Representations via Domain Randomization and Meta-learning [CVPR 2021] (Meta-DR) [153]
A Fourier-Based Framework for Domain Generalization [CVPR 2021] [Code] (FACT) [160]
Open Domain Generalization with Domain-Augmented Meta-Learning [CVPR 2021] [Code] (DAML) [119]
A Simple Feature Augmentation for Domain Generalization [ICCV 2021] (SFA) [142]
Universal Cross-Domain Retrieval Generalizing Across Classes and Domains [ICCV 2021] [Code] (SnMpNet) [150]
Feature Stylization and Domain-aware Contrastive Learning for Domain Generalization [MM 2021] [137]
Adversarial Teacher-Student Representation Learning for Domain Generalization [NeurIPS 2021] [203]
Model-Based Domain Generalization [NeurIPS 2021] [Code] (MBDG) [200]
Optimal Representations for Covariate Shift [ICLR 2022] [Code] (CAD) [223]
Label-Efficient Domain Generalization via Collaborative Exploration and Generalization [MM 2022] [Code] (CEG) [211]

Meta-Learning-Based Methods

Meta-learning-based methods train the model on a meta-train set and improve its performance on a meta-test set for boosting out-of-domain generalization ability.

Learning to Generalize: Meta-Learning for Domain Generalization [AAAI 2018] [Code] (MLDG) [1]
MetaReg: Towards Domain Generalization using Meta-Regularization [NeurIPS 2018] [Code*] (MetaReg) [4]
Feature-Critic Networks for Heterogeneous Domain Generalisation [ICML 2019] [Code] (Feature-Critic) [5]
Episodic Training for Domain Generalization [ICCV 2019] [Code] (Epi-FCR) [7]
Domain Generalization via Model-Agnostic Learning of Semantic Features [NeurIPS 2019] [Code] (MASF) [18]
Domain Generalization via Semi-supervised Meta Learning [arXiv 2020] [Code] (DGSML) [127]
Frustratingly Simple Domain Generalization via Image Stylization [arXiv 2020] [Code] [60]
Domain Generalization for Named Entity Boundary Detection via Metalearning [TNNLS 2020] (METABDRY) [124]
Learning to Learn Single Domain Generalization [CVPR 2020] [Code] (M-ADA) [27]
Learning to Learn with Variational Information Bottleneck for Domain Generalization [ECCV 2020] (MetaVIB) [15]
Sequential Learning for Domain Generalization [ECCV workshop 2020] (S-MLDG) [14]
Shape-Aware Meta-Learning for Generalizing Prostate MRI Segmentation to Unseen Domains [MICCAI 2020] [Code] (SAML) [17]
More is Better: A Novel Multi-view Framework for Domain Generalization [arXiv 2021] [184]
Few-Shot Classification in Unseen Domains by Episodic Meta-Learning Across Visual Domains [ICIP 2021] (x-EML) [180]
Meta-Learned Feature Critics for Domain Generalized Semantic Segmentation [ICIP 2021] [185]
MetaNorm: Learning to Normalize Few-Shot Batches Across Domains [ICLR 2021] [Code] (MetaNorm) [19]
Learning to Generalize Unseen Domains via Memory-based Multi-Source Meta-Learning for Person Re-Identification [CVPR 2021] [Code] (M3L) [12]
Uncertainty-guided Model Generalization to Unseen Domains [CVPR 2021] [Code] [168]
Continual Adaptation of Visual Representations via Domain Randomization and Meta-learning [CVPR 2021] (Meta-DR) [153]
Meta Batch-Instance Normalization for Generalizable Person Re-Identification [CVPR 2021] [Code] (MetaBIN) [13]
Open Domain Generalization with Domain-Augmented Meta-Learning [CVPR 2021] [Code] (DAML) [119]
On Challenges in Unsupervised Domain Generalization [NeurIPS workshop 2021] [178]
Exploiting Domain-Specific Features to Enhance Domain Generalization [NeurIPS 2021] [Code] (mDSDI) [202]

Ensemble Learning-Based Methods

Ensemble learning-based methods mainly train a domain-specific model on each source domain, and then draw on collective wisdom to make accurate prediction.

Exploiting Low-Rank Structure from Latent Domains for Domain Generalization [ECCV 2014] [87]
Visual recognition by learning from web data: A weakly supervised domain generalization approach [CVPR 2015] [89]
Multi-View Domain Generalization for Visual Recognition [ICCV 2015] (MVDG) [88]
Deep Domain Generalization With Structured Low-Rank Constraint [TIP 2017] [91]
Visual Recognition by Learning From Web Data via Weakly Supervised Domain Generalization [TNNLS 2017] [121]
Robust Place Categorization with Deep Domain Generalization [IEEE Robotics and Automation Letters 2018] [Code] (COLD) [97]
Multi-View Domain Generalization Framework for Visual Recognition [TNNLS 2018] [122]
Domain Generalization with Domain-Specific Aggregation Modules [GCPR 2018] (D-SAMs) [92]
Best Sources Forward: Domain Generalization through Source-Specific Nets [ICIP 2018] [90]
Batch Normalization Embeddings for Deep Domain Generalization [arXiv 2020] (BNE) [96]
DoFE: Domain-oriented Feature Embedding for Generalizable Fundus Image Segmentation on Unseen Datasets [TMI 2020] (DoFE) [93]
MS-Net: Multi-Site Network for Improving Prostate Segmentation with Heterogeneous MRI Data [TMI 2020] [Code] (MS-Net) [95]
Generalized Convolutional Forest Networks for Domain Generalization and Visual Recognition [ICLR 2020] (GCFN) [126]
Learning to Optimize Domain Specific Normalization for Domain Generalization [ECCV 2020] (DSON) [94]
Class-conditioned Domain Generalization via Wasserstein Distributional Robust Optimization [ICLR workshop 2021] [175]
Domain and Content Adaptive Convolution for Domain Generalization in Medical Image Segmentation [arXiv 2021] (DCAC) [176]
Dynamically Decoding Source Domain Knowledge for Unseen Domain Generalization [arXiv 2021] (D2SDK) [174]
Domain Adaptive Ensemble Learning [TIP 2021] [Code] (mini-DomainNet dataset) [34]
Generalizable Person Re-identification with Relevance-aware Mixture of Experts [CVPR 2021] (RaMoE) [187]
Learning Transferrable and Interpretable Representations for Domain Generalization [MM 2021] (DTN) [131]
Embracing the Dark Knowledge: Domain Generalization Using Regularized Knowledge Distillation [MM 2021] (KDDG) [157]
TransMatcher: Deep Image Matching Through Transformers for Generalizable Person Re-identification [NeurIPS 2021] [Code] (TransMatcher) [208]

Self-Supervised Learning-Based Methods

Self-supervised learning-based methods improve model generalization by solving some pretext tasks with data itself.

Domain Generalization for Object Recognition with Multi-Task Autoencoders [ICCV 2015] [Code] (MTAE, Rotated MNIST dataset) [6]
Domain Generalization by Solving Jigsaw Puzzles [CVPR 2019] [Code] (JiGen) [98]
Improving Out-Of-Distribution Generalization via Multi-Task Self-Supervised Pretraining [arXiv 2020] [102]
Generalized Convolutional Forest Networks for Domain Generalization and Visual Recognition [ICLR 2020] (GCFN) [126]
Learning from Extrinsic and Intrinsic Supervisions for Domain Generalization [ECCV 2020] [Code] (EISNet) [99]
Zero Shot Domain Generalization [BMVC 2020] [Code] [100]
Out-of-domain Generalization from a Single Source: A Uncertainty Quantification Approach [arXiv 2021] [151]
Self-Supervised Learning Across Domains [TPAMI 2021] [Code] [101]
Multi-Domain Adversarial Feature Generalization for Person Re-Identification [TIP 2021] (MMFA-AAE) [144]
Scale Invariant Domain Generalization Image Recapture Detection [ICONIP 2021] (SADG) [177]
Domain Generalisation with Domain Augmented Supervised Contrastive Learning [AAAI Student Abstract 2021]
Progressive Domain Expansion Network for Single Domain Generalization [CVPR 2021] [Code] (PDEN) [141]
FedDG: Federated Domain Generalization on Medical Image Segmentation via Episodic Learning in Continuous Frequency Space [CVPR 2021] [Code] (FedDG) [147]
Boosting the Generalization Capability in Cross-Domain Few-shot Learning via Noise-enhanced Supervised Autoencoder [ICCV 2021] (NSAE) [194]
A Style and Semantic Memory Mechanism for Domain Generalization [ICCV 2021] (STEAM) [130]
SelfReg: Self-Supervised Contrastive Regularization for Domain Generalization [ICCV 2021] (SelfReg) [138]
Domain Generalization for Mammography Detection via Multi-style and Multi-view Contrastive Learning [MICCAI 2021] [Code] (MSVCL) [172]
Feature Stylization and Domain-aware Contrastive Learning for Domain Generalization [MM 2021] [137]
Adversarial Teacher-Student Representation Learning for Domain Generalization [NeurIPS 2021]
Domain Generalization via Contrastive Causal Learning [arXiv 2022] (CCM) [212]
Towards Unsupervised Domain Generalization [CVPR 2022] (DARLING) [69]
Unsupervised Domain Generalization by Learning a Bridge Across Domains [CVPR 2022] [Code] (BrAD) [182]

Disentangled Representation Learning-Based Methods

Disentangled representation learning-based methods aim to disentangle domain-specific and domain-invariant parts from source data, and then adopt the domain-invariant one for inference on the target domains.

Undoing the Damage of Dataset Bias [ECCV 2012] [Code] [103]
Deeper, Broader and Artier Domain Generalization [ICCV 2017] [Code] [2]
DIVA: Domain Invariant Variational Autoencoders [ICML workshop 2019] [Code] (DIVA) [107]
Efficient Domain Generalization via Common-Specific Low-Rank Decomposition [ICML 2020] [Code] (CSD) [105]
Cross-Domain Face Presentation Attack Detection via Multi-Domain Disentangled Representation Learning [CVPR 2020] [106]
Learning to Balance Specificity and Invariance for In and Out of* Domain Generalization [ECCV 2020] [Code] (DMG) [104]
Towards Principled Disentanglement for Domain Generalization [arXiv 2021] [Code] (DDG) [170]
Meta-Learned Feature Critics for Domain Generalized Semantic Segmentation [ICIP 2021] [185]
DecAug: Out-of-Distribution Generalization via Decomposed Feature Representation and Semantic Augmentation [AAAI 2021] [Code] (DecAug) [171]
Robustnet: Improving Domain Generalization in Urban-Scene Segmentation via Instance Selective Whitening [CVPR 2021] [Code] (RobustNet) [193]
Reducing Domain Gap by Reducing Style Bias [CVPR 2021] [Code] (SagNet) [230]
Shape-Biased Domain Generalization via Shock Graph Embeddings [ICCV 2021] [149]
Domain-Invariant Disentangled Network for Generalizable Object Detection [ICCV 2021] [143]
Domain Generalization via Feature Variation Decorrelation [MM 2021] [146]
Exploiting Domain-Specific Features to Enhance Domain Generalization [NeurIPS 2021] [Code] (mDSDI) [202]
Variational Disentanglement for Domain Generalization [TMLR 2022] (VDN) [210]
Intra-Source Style Augmentation for Improved Domain Generalization [WACV 2023] (ISSA) [215]

Regularization-Based Methods

Regularization-based methods leverage regularization terms to prevent the overfitting, or design optimization strategies to guide the training.

Generalizing from Several Related Classification Tasks to a New Unlabeled Sample [NeurIPS 2011] [113]
MetaReg: Towards Domain Generalization using Meta-Regularization [NeurIPS 2018] [Code*] (MetaReg) [4]
Invariant Risk Minimization [arXiv 2019] [Code] (IRM, Colored MNIST dataset) [165]
Learning Robust Representations by Projecting Superficial Statistics Out [ICLR 2019] [Code] (HEX, ImageNet-Sketch dataset) [35]
Distributionally Robust Neural Networks for Group Shifts On the Importance of Regularization for Worst-Case Generalization [ICLR 2020] [Code] (DroupDRO) [218]
Self-challenging Improves Cross-Domain Generalization [ECCV 2020] [Code] (RSC) [64]
Energy-based Out-of-distribution Detection [NeurIPS 2020] [Code] [181]
When Can We Formulate the Out-of-Distribution Generalization Problem as an Invariance Problem? [arXiv 2021] [Code*] (IGA) [219]
Learning Representations that Support Robust Transfer of Predictors [arXiv 2021] [Code] (TRM) [220]
SAND-mask: An Enhanced Gradient Masking Strategy for the Discovery of Invariances in Domain Generalization [arXiv 2021] [Code*] (SANDMask) [222]
Out-of-Distribution Generalization via Risk Extrapolation [ICML 2021] (VREx) [190]
Learning Explanations that are Hard to Vary [ICLR 2021] [Code*] (ANDMask) [221]
A Fourier-Based Framework for Domain Generalization [CVPR 2021] [Code] (FACT) [160]
Domain Generalization via Gradient Surgery [ICCV 2021] [Code] (Agr) [148]
SelfReg: Self-Supervised Contrastive Regularization for Domain Generalization [ICCV 2021] (SelfReg) [138]
Embracing the Dark Knowledge: Domain Generalization Using Regularized Knowledge Distillation [MM 2021]
Model-Based Domain Generalization [NeurIPS 2021] [Code] (MBDG) [200]
Swad: Domain Generalization by Seeking Flat Minima [NeurIPS 2021] [Code] (SWAD) [201]
Training for the Future: A Simple Gradient Interpolation Loss to Generalize Along Time [NeurIPS 2021] [Code] (GI) [204]
Adaptive Risk Minimization: Learning to Adapt to Domain Shift [NeurIPS 2021] [Code] (ARM) [228]
Gradient Starvation: A Learning Proclivity in Neural Networks [NeurIPS 2021] [Code*] (SD) [225]
Quantifying and Improving Transferability in Domain Generalization [NeurIPS 2021] [Code] [206]
Gradient Matching for Domain Generalization [ICLR 2022] [Code] (Fish) [224]
Fishr: Invariant Gradient Variances for Our-of-distribution Generalization [ICML 2022] [Code] (Fishr) [173]

Normalization-Based Methods

Normalization-based methods calibrate data from different domains by normalizing them with their statistic.

Deep CORAL: Correlation Alignment for Deep Domain Adaptation [ECCV 2016] [Code] (CORAL) [229]
Batch Normalization Embeddings for Deep Domain Generalization [arXiv 2020] (BNE) [96]
Learning to Optimize Domain Specific Normalization for Domain Generalization [ECCV 2020] (DSON) [94]
MetaNorm: Learning to Normalize Few-Shot Batches Across Domains [ICLR 2021] [Code] (MetaNorm) [19]
Meta Batch-Instance Normalization for Generalizable Person Re-Identification [CVPR 2021] [Code] (MetaBIN) [13]
- Learning to Generalize Unseen Domains via Memory-based Multi-Source Meta-Learning for Person Re-Identification [CVPR 2021] [Code] (M3L) [12]
Adversarially Adaptive Normalization for Single Domain Generalization [CVPR 2021] (ASR) [116]
Collaborative Optimization and Aggregation for Decentralized Domain Generalization and Adaptation [ICCV 2021] (COPDA) [159]
Domain Generalization through Audio-Visual Relative Norm Alignment in First Person Action Recognition [WACV 2022] (RNA-Net) [186]

Information-Based Methods

Information-based methods utilize techniques of information theory to realize domain generalization.

Learning to Learn with Variational Information Bottleneck for Domain Generalization [ECCV 2020] (MetaVIB) [15]
Progressive Domain Expansion Network for Single Domain Generalization [CVPR 2021] [Code] (PDEN) [141]
Learning To Diversify for Single Domain Generalization [ICCV 2021] [Code] [158]
Invariance Principle Meets Information Bottleneck for Out-Of-Distribution Generalization [NeurIPS 2021] [Code] (IB-IRM) [207]
Exploiting Domain-Specific Features to Enhance Domain Generalization [NeurIPS 2021] [Code] (mDSDI) [202]
Invariant Information Bottleneck for Domain Generalization [AAAI 2022] [Code] (IIB) [140]

Causality-Based Methods

Causality-based methods analyze and address the domain generalization problem from a causal perspective.

Invariant Risk Minimization [arXiv 2019] [Code] (IRM, Colored MNIST dataset) [165]
Learning Domain-Invariant Relationship with Instrumental Variable for Domain Generalization [arXiv 2021] (IV-DG) [163]
A Causal Framework for Distribution Generalization [TPAMI 2021] [Code] (NILE) [191]
Domain Generalization using Causal Matching [ICML 2021] [Code] (MatchDG) [73]
Deep Stable Learning for Out-of-Distribution Generalization [CVPR 2021] [Code] (StableNet) [117]
Out-of-Distribution Generalization via Risk Extrapolation [ICML 2021] [Code] (VREx) [217]
A Style and Semantic Memory Mechanism for Domain Generalization [ICCV 2021] (STEAM) [130]
Learning Causal Semantic Representation for Out-of-Distribution Prediction [NeurIPS 2021] [Code] (CSG-ind) [145]
Recovering Latent Causal Factor for Generalization to Distributional Shifts [NeurIPS 2021] [Code] (LaCIM) [152]
On Calibration and Out-of-domain Generalization [NeurIPS 2021]
Invariance Principle Meets Information Bottleneck for Out-Of-Distribution Generalization [NeurIPS 2021] [Code] (IB-ERM, IB-IRM) [207]
Domain Generalization via Contrastive Causal Learning [arXiv 2022] (CCM) [212]
Invariant Causal Mechanisms through Distribution Matching [arXiv 2022] [Code*] (CausIRL-CORAL, CausIRL-MMD) [216]
Invariant Information Bottleneck for Domain Generalization [AAAI 2022] [Code] (IIB) [140]

Inference-Time-Based Methods

Inference-time-based methods leverage the unlabeled target data, which is available at inference-time, to improve generalization performance without further model training.

Generalization on Unseen Domains via Inference-Time Label-Preserving Target Projections [CVPR 2021] [Code] [118]
Adaptive Methods for Real-World Domain Generalization [CVPR 2021] [Code] (DA-ERM) [132]
Test-Time Classifier Adjustment Module for Model-Agnostic Domain Generalization [NeurIPS 2021] [Code] (T3A) [136]

Neural Architecture Search-based Methods

Neural architecture search-based methods aim to dynamically tune the network architecture to improve out-of-domain generalization.

NAS-OoD Neural Architecture Search for Out-of-Distribution Generalization [ICCV 2021] (NAS-OoD) [129]

Single Domain Generalization

The goal of single domain generalization task is to improve model performance on unknown target domains by using data from only one source domain.

Learning to Learn Single Domain Generalization [CVPR 2020] [Code] (M-ADA) [27]
Out-of-domain Generalization from a Single Source: A Uncertainty Quantification Approach [arXiv 2021] [151]
Uncertainty-guided Model Generalization to Unseen Domains [CVPR 2021] [Code] [168]
Adversarially Adaptive Normalization for Single Domain Generalization [CVPR 2021] (ASR) [116]
Progressive Domain Expansion Network for Single Domain Generalization [CVPR 2021] [Code] (PDEN) [141]
Learning To Diversify for Single Domain Generalization [ICCV 2021] [Code] [158]
Intra-Source Style Augmentation for Improved Domain Generalization [WACV 2023] (ISSA) [215]

Semi/Weak/Un-Supervised Domain Generalization

Semi/weak-supervised domain generalization assumes that a part of the source data is unlabeled, while unsupervised domain generalization assumes no training supervision.

Visual recognition by learning from web data: A weakly supervised domain generalization approach [CVPR 2015] [89]
Visual Recognition by Learning From Web Data via Weakly Supervised Domain Generalization [TNNLS 2017] [121]
Domain Generalization via Semi-supervised Meta Learning [arXiv 2020] [Code] (DGSML) [127]
Deep Semi-supervised Domain Generalization Network for Rotary Machinery Fault Diagnosis under Variable Speed [IEEE Transactions on Instrumentation and Measurement 2020] (DSDGN) [125]
Semi-Supervised Domain Generalization with Stochastic StyleMatch [arXiv 2021] [Code] (StyleMatch) [54]
Better Pseudo-label Joint Domain-aware Label and Dual-classifier for Semi-supervised Domain Generalization [arXiv 2021] [156]
Semi-Supervised Domain Generalization in RealWorld: New Benchmark and Strong Baseline [arXiv 2021] [179]
On Challenges in Unsupervised Domain Generalization [NeurIPS workshop 2021] [178]
Domain-Specific Bias Filtering for Single Labeled Domain Generalization [IJCV 2022] [Code] (DSBF) [162]
Towards Unsupervised Domain Generalization [CVPR 2022] (DARLING) [69]
Unsupervised Domain Generalization by Learning a Bridge Across Domains [CVPR 2022] [Code] (BrAD) [182]
Label-Efficient Domain Generalization via Collaborative Exploration and Generalization [MM 2022] [Code] (CEG) [211]

Open/Heterogeneous Domain Generalization

Open/heterogeneous domain generalization assumes the label space of one domain is different from that of another domain.

Feature-Critic Networks for Heterogeneous Domain Generalisation [ICML 2019] [Code] (Feature-Critic) [5]
Episodic Training for Domain Generalization [ICCV 2019] [Code] (Epi-FCR) [7]
Towards Recognizing Unseen Categories in Unseen Domains [ECCV 2020] [Code] (CuMix) [57]
Heterogeneous Domain Generalization via Domain Mixup [ICASSP 2020] [Code] [128]
Open Domain Generalization with Domain-Augmented Meta-Learning [CVPR 2021] [Code] (DAML) [119]
Universal Cross-Domain Retrieval Generalizing Across Classes and Domains [ICCV 2021] [Code] (SnMpNet) [150]

Federated Domain Generalization

Federated domain generalization assumes that source data is distributed and can not be fused for data privacy protection.

Collaborative Semantic Aggregation and Calibration for Separated Domain Generalization [arXiv 2021] [Code] (CSAC) [161]
FedDG: Federated Domain Generalization on Medical Image Segmentation via Episodic Learning in Continuous Frequency Space [CVPR 2021] [Code] (FedDG) [147]
Collaborative Optimization and Aggregation for Decentralized Domain Generalization and Adaptation [ICCV 2021] (COPDA) [159]

Applications

Person Re-Identification

Deep Domain-Adversarial Image Generation for Domain Generalisation [AAAI 2020] [Code]
Learning to Generate Novel Domains for Domain Generalization [ECCV 2020] [Code] (L2A-OT, Digits-DG dataset) [28]
Learning Generalisable Omni-Scale Representations for Person Re-Identification [TPAMI 2021] [Code] [114]
Multi-Domain Adversarial Feature Generalization for Person Re-Identification [TIP 2021] (MMFA-AAE) [144]
Domain Generalization with Mixstyle [ICLR 2021] [Code] (MixStyle) [56]
Learning to Generalize Unseen Domains via Memory-based Multi-Source Meta-Learning for Person Re-Identification [CVPR 2021] [Code] (M3L) [12]
Meta Batch-Instance Normalization for Generalizable Person Re-Identification [CVPR 2021] [Code] (MetaBIN) [13]
Generalizable Person Re-identification with Relevance-aware Mixture of Experts [CVPR 2021] (RaMoE) [187]
TransMatcher: Deep Image Matching Through Transformers for Generalizable Person Re-identification [NeurIPS 2021] [Code] (TransMatcher) [208]

Face Recognition & Anti-Spoofing

Multi-Adversarial Discriminative Deep Domain Generalization for Face Presentation Attack Detection [CVPR 2019] [Code] (MADDG) [78]
Towards Universal Representation Learning for Deep Face Recognition [CVPR 2020] [Code] [22]
Cross-Domain Face Presentation Attack Detection via Multi-Domain Disentangled Representation Learning [CVPR 2020] [106]
Single-Side Domain Generalization for Face Anti-Spoofing [CVPR 2020] [Code] (SSDG) [79]

Publications

Top Conference	Papers
before 2014	CVPR: [8], [11]; ICCV: [16], [41]; NeurIPS: [31], [113]; ECCV: [32], [87], [103]; ICML: [65]
2015	CVPR: [89]; ICML: [30]; ICCV: [6], [46], [88]
2016	CVPR: [42], [44], [120]; IJCAI: [66]; ECCV: [43], [47], [229]
2017	CVPR: [20]; ICCV: [2], [71]; NeurIPS: [38]
2018	ICLR: [1], [68]; ICLR: [53]; CVPR: [76]; ECCV: [45], [77]; NeurIPS: [4], [25]
2019	ICLR: [35], [37]; CVPR: [78], [98]; ICML: [5], [107], [110]; ICCV: [7], [21], [33], [62], [63]; NeurIPS: [18]
2020	ICLR: [55], [83], [218]; ICLR: [126]; CVPR: [22], [27], [79], [106]; ICML: [105]; ECCV: [14], [15], [28], [57], [64], [94], [99], [104]; NeurIPS: [75], [86], [112], [181]
2021	ICLR: [19], [56], [59], [134], [175], [196]; ICLR: [139], [171], [221]; CVPR: [12], [13], [115], [116], [117], [118], [119], [132], [141], [147], [153], [160], [168], [187], [193]; IJCAI: [155], [195], [230]; ICML: [73], [190], [217]; ICCV: [129], [130], [133], [135], [138], [142], [143], [148], [149], [150], [158], [159], [194]; MM: [131], [137], [146], [157]; NeurIPS: [136], [145], [152], [154], [198], [199], [200], [201], [202], [203], [204], [205], [206], [207], [208], [228], [225]
2022	AAAI: [140]; ICLR: [213], [224]; CVPR: [69], [182], [214]; ICML: [173]; MM: [211]
2023	WACV: [215]; ICLR: [223]

Top Journal	Papers
before 2017	IJCV: [9], [10]; JMLR: [226]
2017	TPAMI: [67]; TIP: [91]
2021	TIP: [34], [144]; TPAMI: [101], [114], [191], [197]; JMLR: [188]
2022	TMLR: [209], [210]; IJCV: [162]

arXiv	Papers
before 2014	[40]
2017	[36], [52]
2018	[166]
2019	[81], [123], [165], [169]
2020	[60], [82], [96], [102], [127], [189], [227]
2021	[3], [54], [58], [151], [156], [161], [163], [170], [174], [176], [178], [179], [184], [192], [219], [222]
2022	[183], [212], [216], [220]

Else	Papers
before 2018	[29], [39], [48], [49], [50], [51], [90], [92], [97], [109], [111], [121], [122]
2019	[26], [72], [84], [167]
2020	[17], [23], [24], [61], [70], [74], [80], [85], [93], [95], [100], [124], [125], [128], [164]
2021	[108], [172], [177], [180], [185]
2022	[186]

Datasets

Evaluations on the following datasets often follow leave-one-domain-out protocol: randomly choose one domain to hold out as the target domain, while the others are used as the source domain(s).

Datasets (download link)	Description	Related papers
Colored MNIST [165]	Handwritten digit recognition; 3 domains: {0.1, 0.3, 0.9}; 70,000 samples of dimension (2, 28, 28); 2 classes	[82], [138], [140], [149], [152], [154], [165], [171], [173], [190], [200], [202], [214], [216], [217], [219], [220], [222], [224]
Rotated MNIST [6] (original)	Handwritten digit recognition; 6 domains with rotated degree: {0, 15, 30, 45, 60, 75}; 7,000 samples of dimension (1, 28, 28); 10 classes	[5], [6], [15], [35], [53], [55], [63], [71], [73], [74], [76], [77], [86], [90], [105], [107], [138], [140], [170], [173], [202], [204], [206], [216], [222], [224]
Digits-DG [28]	Handwritten digit recognition; 4 domains: {MNIST [29], MNIST-M [30], SVHN [31], SYN [30]}; 24,000 samples; 10 classes	[21], [25], [27], [28], [34], [35], [55], [59], [63], [94], [98], [116], [118], [130], [141], [142], [146], [151], [153], [157], [158], [159], [160], [166], [168], [179], [189], [203], [209], [210]
VLCS [16] (1; or original)	Object recognition; 4 domains: {Caltech [8], LabelMe [9], PASCAL [10], SUN [11]}; 10,729 samples of dimension (3, 224, 224); 5 classes; about 3.6 GB	[2], [6], [7], [14], [15], [18], [60], [61], [64], [67], [68], [70], [71], [74], [76], [77], [81], [83], [86], [91], [98], [99], [101], [102], [103], [117], [118], [126], [127], [131], [132], [136], [138], [140], [142], [145], [146], [148], [149], [161], [170], [173], [174], [184], [190], [195], [199], [201], [202], [203], [209], [216], [217], [222], [223], [224]
Office31+Caltech [32] (1)	Object recognition; 4 domains: {Amazon, Webcam, DSLR, Caltech}; 4,652 samples in 31 classes (office31) or 2,533 samples in 10 classes (office31+caltech); 51 MB	[6], [35], [67], [68], [70], [71], [80], [91], [96], [119], [131], [167]
OfficeHome [20] (1; or original)	Object recognition; 4 domains: {Art, Clipart, Product, Real World}; 15,588 samples of dimension (3, 224, 224); 65 classes; 1.1 GB	[19], [54], [28], [34], [55], [58], [60], [61], [64], [80], [92], [94], [98], [101], [118], [126], [130], [131], [132], [133], [137], [138], [140], [146], [148], [156], [159], [160], [162], [163], [167], [173], [174], [178], [179], [182], [184], [189], [190], [199], [201], [202], [203], [206], [211], [212], [214], [216], [217], [220], [222], [223], [224], [230]
PACS [2] (1; or original)	Object recognition; 4 domains: {photo, art_painting, cartoon, sketch}; 9,991 samples of dimension (3, 224, 224); 7 classes; 174 MB	[1], [2], [4], [5], [14], [15], [18], [19], [34], [54], [28], [35], [55], [56], [57], [58], [59], [60], [61], [64], [69], [73], [77], [80], [81], [82], [83], [84], [86], [90], [92], [94], [96], [98], [99], [101], [102], [104], [105], [116], [117], [118], [127], [129], [130], [131], [132], [136], [137], [138], [139], [140], [142], [145], [146], [148], [149], [153], [156], [157], [158], [159], [160], [161], [162], [163], [167], [170], [171], [173], [174], [178], [179], [180], [182], [184], [189], [190], [195], [199], [200], [201], [202], [203], [206], [209], [210], [211], [212], [214], [216], [217], [220], [222], [223], [224], [230]
DomainNet [33] (clipart, infograph, painting, quick-draw, real, and sketch; or original)	Object recognition; 6 domains: {clipart, infograph, painting, quick-draw, real, sketch}; 586,575 samples of dimension (3, 224, 224); 345 classes; 1.2 GB + 4.0 GB + 3.4 GB + 439 MB + 5.6 GB + 2.5 GB	[34], [57], [69], [104], [119], [130], [131], [132], [133], [138], [140], [150], [173], [182], [189], [201], [202], [203], [216], [222], [223], [224], [230]
mini-DomainNet [34]	Object recognition; a smaller and less noisy version of DomainNet; 4 domains: {clipart, painting, real, sketch}; 140,006 samples	[34], [130], [156], [157], [210]
ImageNet-Sketch [35]	Object recognition; 2 domains: {real, sketch}; 50,000 samples	[64]
VisDA-17 [36]	Object recognition; 3 domains of synthetic-to-real generalization; 280,157 samples	[119], [182]
CIFAR-10-C / CIFAR-100-C / ImageNet-C [37] (original)	Object recognition; the test data are damaged by 15 corruptions (each with 5 intensity levels) drawn from 4 categories (noise, blur, weather, and digital); 60,000/60,000/1.3M samples	[27], [69], [74], [116], [141], [151], [168]
Visual Decathlon (VD) [38]	Object/action/handwritten/digit recognition; 10 domains from the combination of 10 datasets; 1,659,142 samples	[5], [7], [128]
IXMAS [39]	Action recognition; 5 domains with 5 camera views, 10 subjects, and 5 actions; 1,650 samples	[7], [14], [67], [76]
SYNTHIA [42]	Semantic segmentation; 15 domains with 4 locations and 5 weather conditions; 2,700 samples	[27], [62], [115], [141], [151], [185], [193]
GTA5-Cityscapes [43], [44]	Semantic segmentation; 2 domains of synthetic-to-real generalization; 29,966 samples	[62], [115], [185], [193]
Cityscapes-ACDC [44] (original)	Semantic segmentation; real life domain shifts, ACDC contains four different weather conditions: rain, fog, snow, night	[215]
Terra Incognita (TerraInc) [45] (1 and 2; or original)	Animal classification; 4 domains captured at different geographical locations: {L100, L38, L43, L46}; 24,788 samples of dimension (3, 224, 224); 10 classes; 6.0 GB + 8.6 MB	[132], [136], [138], [140], [173], [201], [202], [207], [212], [214], [216], [222], [223], [224]
Market-Duke [46], [47]	Person re-idetification; cross-dataset re-ID; heterogeneous DG with 2 domains; 69,079 samples	[12], [13], [28], [55], [56], [58], [114], [144], [187], [208]

Libraries

We list the GitHub libraries of domain generalization (sorted by stars).

DeepDG (jindongwang): Deep Domain Generalization Toolkit.
Transfer Learning Library (thuml) for Domain Adaptation, Task Adaptation, and Domain Generalization.
DomainBed (facebookresearch) [134] is a suite to test domain generalization algorithms.
Dassl (KaiyangZhou): A PyTorch toolbox for domain adaptation, semi-supervised learning, and domain generalization.

Lectures & Tutorials & Talks

(Talk 2021) Generalizing to Unseen Domains: A Survey on Domain Generalization [155]. [Video] [Slides] (Jindong Wang (MSRA), in Chinese)

Other Resources

A collection of domain generalization papers organized by amber0309.
A collection of domain generalization papers organized by jindongwang.
A collection of papers on domain generalization, domain adaptation, causality, robustness, prompt, optimization, generative model, etc, organized by yfzhang114.
Adaptation and Generalization Across Domains in Visual Recognition with Deep Neural Networks [PhD 2020, Kaiyang Zhou (University of Surrey)] [164]

Contributing & Contact

Feel free to contribute to our repository.

If you woulk like to correct mistakes, please do it directly;
If you would like to add/update papers, please finish the following tasks (if necessary):
1. Find the max index (current max: [230], not used: none), and create a new one.
2. Update Publications.
3. Update Papers.
4. Update Datasets.
If you have any questions or advice, please contact us by email (yuanjk@zju.edu.cn) or GitHub issues.

Thank you for your cooperation and contributions!

Acknowledgements

The designed hierarchy of the Contents is mainly based on awesome-domain-adaptation.

We refer to [3] to design the Contents and the table of Datasets.

junkunyuan / Awesome-Domain-Generalization