FedHUG: Federated Heterogeneous Unsupervised Generalization for Remote Physiological Measurements

TL;DR

This work introduces Federated Unsupervised Domain Generalization (FUDG) for remote physiological measurement and proposes the FedHUG framework, designed to generalize across unseen domains using unlabeled local data. It combines Minimal Bias Aggregation (MBA) to weight client contributions by bias and Global Distribution Aware Learning Controller (GDLC) to model and leverage a global semantic distribution, enabling tail-aware learning in a privacy-preserving federated setting. The approach supports both RGB video and mmWave RPM modalities, achieves superior generalization over state-of-the-art federated methods, and includes extensive ablations and visual analyses to illustrate bias elimination and tail learning. The work advances practical deployment of RPM in real-world, privacy-sensitive environments by reducing domain bias and handling long-tail label distributions.

Abstract

Remote physiological measurement gained wide attention, while it requires collecting users' privacy-sensitive information, and existing contactless measurements still rely on labeled client data. This presents challenges when we want to further update real-world deployed models with numerous user data lacking labels. To resolve these challenges, we instantiate a new protocol called Federated Unsupervised Domain Generalization (FUDG) in this work. Subsequently, the \textbf{Fed}erated \textbf{H}eterogeneous \textbf{U}nsupervised \textbf{G}eneralization (\textbf{FedHUG}) framework is proposed and consists of: (1) Minimal Bias Aggregation module dynamically adjusts aggregation weights based on prior-driven bias evaluation to cope with heterogeneous non-IID features from multiple domains. (2) The Global Distribution-aware Learning Controller parameterizes the label distribution and dynamically manipulates client-specific training strategies, thereby mitigating the server-client label distribution skew and long-tail issue. The proposal shows superior performance across state-of-the-art techniques in estimation with either RGB video or mmWave radar. The code will be released.

Figures (7)

• Figure 1: Overview of FUDG. We take the video-based physiological estimation as an example.
• Figure 2: Illustration of three different aim domains. Here, the dashed circle’s center represents the semantic features, and the closer a domain is to the center, the less interference it contains.
• Figure 3: Illustration of the self-similarity matrix of the original data and the augmented data. The intensity of the brightness in the matrix reflects the similarity between time windows.
• Figure 4: Illustration of the distribution of $s$ in different datasets (i.e., PURE song2021pulsegan, BUAA xi2020image, VIPL lu2021dual, MMPD tang2023mmpd).
• Figure 5: Illustration of the HR distribution in different datasets.