This repository is a curated collection of resources on domain generalization (DG), a machine learning field focused on building models that perform well on unseen target domains. It targets researchers and practitioners in machine learning, computer vision, and related areas, providing a comprehensive overview of papers, datasets, and libraries.

How It Works

The repository categorizes DG methods into distinct approaches, including domain alignment, data augmentation, meta-learning, ensemble learning, self-supervised learning, disentangled representation learning, regularization, normalization, information-based, causality-based, and neural architecture search. It also covers specific DG scenarios like single-domain, semi/weak/un-supervised, open/heterogeneous, and federated DG. The core value lies in its extensive, categorized bibliography of research papers, often linking to implementations and providing publication details.

Quick Start & Requirements

This repository is a curated list and does not have a direct installation or execution command. It serves as a knowledge base.

Highlighted Details

• Comprehensive categorization of domain generalization techniques and research papers.
• Extensive list of datasets commonly used for evaluating DG methods, with links and descriptions.
• Links to relevant libraries and toolkits for domain generalization research.
• Includes a timeline of publications in top conferences and journals.

Maintenance & Community

The repository is actively maintained, with recent updates including papers from WACV 2025 and ICCV 2023. Contributions are welcomed via GitHub issues or email.

Licensing & Compatibility

The repository itself is a collection of links and information; it does not host code or data directly, thus licensing is not applicable to the repository's content. Individual papers and code repositories linked within will have their own licenses.

Limitations & Caveats

While comprehensive, the repository is a bibliography and does not provide executable code or pre-trained models. Users must refer to individual paper links for implementation details and potential usage. Some code links may be unofficial or require specific environments.