FedAlign: Federated Domain Generalization with Cross-Client Feature Alignment

Authors: Sunny Gupta, Vinay Sutar, Varunav Singh, Amit Sethi
Conference: CVPR 2025
Paper: (https://openaccess.thecvf.com/content/CVPR2025W/FedVision/html/Gupta_FedAlign_Federated_Domain_Generalization_with_Cross-Client_Feature_Alignment_CVPRW_2025_paper.html)

---

Overview

FedAlign is a lightweight, privacy-preserving framework for Federated Domain Generalization (FedDG) that enhances generalization to unseen domains without compromising data privacy. It addresses key challenges such as:

• Strict privacy constraints in Federated Learning (FL)
• Non-i.i.d. data distributions across clients
• Limited domain diversity

Key Contributions

• Cross-Client Feature Extension Module
  Enhances local domain representation using domain-invariant feature perturbation and selective cross-client feature transfer.

• Dual-Stage Alignment Module
  Aligns both feature embeddings and predictions across clients to extract robust, domain-invariant representations.

• Lightweight & Scalable
  Requires minimal computation and communication overhead, making it suitable for real-world federated setups.

---

Supported datasets include:

• MiniDomainNet
• PACS
• OfficeHome

---

How to Run

You can run the training script using the following command:

python main.py [Algorithm] -d [Dataset Name] [other arguments]

python main.py FedAlign -d minidomainnet

If you find this work useful, please cite as

@inproceedings{gupta2025fedalign,
  title={FedAlign: Federated Domain Generalization with Cross-Client Feature Alignment},
  author={Gupta, Sunny and Sutar, Vinay and Singh, Varunav and Sethi, Amit},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={1801--1810},
  year={2025}
}