Parameter-Efficient Domain Adaption for CSI Crowd-Counting via Self-Supervised Learning with Adapter Modules

TL;DR

This work tackles domain shift in WiFi CSI crowd-counting by marrying self-supervised contrastive learning with parameter-efficient adapter fine-tuning in a two-stage framework. A CSI-ResNet-A encoder, enhanced with Squeeze-and-Excitation blocks and lightweight adapters, is pre-trained on unlabelled CSI data and then adapted to new environments with minimal labelled data, while a stateful counting module converts class predictions into stable occupancy estimates. The approach yields a high-accuracy, data-efficient solution, achieving a final occupancy MAE of $0.44$ in 10-shot WiFlow experiments and near-perfect Generalisation Index values, with a new state-of-the-art $98.8\%$ accuracy on the WiAR benchmark. Ablation studies demonstrate that adapters provide substantial parameter savings (up to $97.2\%$ fewer trainable parameters) with minimal performance loss, and SE blocks are crucial when adaptation capacity is limited, highlighting the framework's practicality for real-world, privacy-preserving IoT deployments.

Abstract

Device-free crowd-counting using WiFi Channel State Information (CSI) is a key enabling technology for a new generation of privacy-preserving Internet of Things (IoT) applications. However, practical deployment is severely hampered by the domain shift problem, where models trained in one environment fail to generalise to another. To overcome this, we propose a novel two-stage framework centred on a CSI-ResNet-A architecture. This model is pre-trained via self-supervised contrastive learning to learn domain-invariant representations and leverages lightweight Adapter modules for highly efficient fine-tuning. The resulting event sequence is then processed by a stateful counting machine to produce a final, stable occupancy estimate. We validate our framework extensively. On our WiFlow dataset, our unsupervised approach excels in a 10-shot learning scenario, achieving a final Mean Absolute Error (MAE) of just 0.44--a task where supervised baselines fail. To formally quantify robustness, we introduce the Generalisation Index (GI), on which our model scores near-perfectly, confirming its ability to generalise. Furthermore, our framework sets a new state-of-the-art public WiAR benchmark with 98.8\% accuracy. Our ablation studies reveal the core strength of our design: adapter-based fine-tuning achieves performance within 1\% of a full fine-tune (98.84\% vs. 99.67\%) while training 97.2\% fewer parameters. Our work provides a practical and scalable solution for developing robust sensing systems ready for real-world IoT deployments.

Figures (11)

• Figure 1: Overview of the proposed system pipeline.
• Figure 2: A raw CSI waveform shown before and after denoising with a 4th-order Butterworth filter.
• Figure 3: The proposed CSI-ResNet-A architecture.
• Figure 4: t-SNE visualization of the pre-trained CSI embeddings.
• Figure 5: State machine for robust occupancy counting.