EMG-UP: Unsupervised Personalization in Cross-User EMG Gesture Recognition

TL;DR

EMG-UP tackles cross-user EMG gesture recognition by offering a source-free, unsupervised personalization pipeline. It combines Sequence-Cross Perspective Contrastive Learning to extract robust, user-specific representations with Pseudo-Label-Guided Fine-Tuning to tailor models to individuals without accessing source data. The method demonstrates state-of-the-art performance on emg2pose and SH2024 datasets, outperforming strong baselines by notable margins and preserving privacy. This work enables scalable, per-user EMG interfaces suitable for real-world deployment, with future work aimed at real-time, in-the-wild adaptation. Overall, EMG-UP advances personalized EMG interaction by eliminating source-data requirements while maintaining strong generalization across users.

Abstract

Cross-user electromyography (EMG)-based gesture recognition represents a fundamental challenge in achieving scalable and personalized human-machine interaction within real-world applications. Despite extensive efforts, existing methodologies struggle to generalize effectively across users due to the intrinsic biological variability of EMG signals, resulting from anatomical heterogeneity and diverse task execution styles. To address this limitation, we introduce EMG-UP, a novel and effective framework for Unsupervised Personalization in cross-user gesture recognition. The proposed framework leverages a two-stage adaptation strategy: (1) Sequence-Cross Perspective Contrastive Learning, designed to disentangle robust and user-specific feature representations by capturing intrinsic signal patterns invariant to inter-user variability, and (2) Pseudo-Label-Guided Fine-Tuning, which enables model refinement for individual users without necessitating access to source domain data. Extensive evaluations show that EMG-UP achieves state-of-the-art performance, outperforming prior methods by at least 2.0% in accuracy.

Figures (4)

• Figure 1: Conventional domain-adaptation pipelines vs. our EMG-UP personalization. Previous methods rely on Domain Adaptation to obtain an adapted model, while our proposed approach focuses on personalization to derive a private, user-specific model. Both publicly available dataset and a private dataset is used to evaluate our framework.
• Figure 2: The framework of our method. Our method consists of two stages: (1)Sequence-Cross Perspective Contrastive Learning, and (2)Pseudo-Label-Guided Fine-Tuning.
• Figure 3: Illustration of sequence-cross perspective contrastive learning. This module captures bidirectional relationships in EMG signal sequences through augmented views.
• Figure 4: Two Alignment Steps of EMG-UP Framework. This figure illustrates the two-stage aligment scheme of the proposed framework: (1) Individual-Specific Distribution Alignment in Sequence-Cross Perspective Contrastive Learning. (2) Individual-Specific Personalized Alignment in Pseudo-Label-Guided Fine-Tuning.