Vision Reimagined: AI-Powered Breakthroughs in WiFi Indoor Imaging

TL;DR

The paper reframes WiFi indoor imaging under phaseless inverse scattering as a cross-modal image generation task and introduces WiFi-GEN, a three-component generative model (WiFi Signal Encoder, Hierarchical Control Signal Network, and GAN-like WiFi Generator) trained with L2 and LPIPS losses. Trained on a large 80k synthetic dataset and validated with both simulations and physical experiments, WiFi-GEN achieves a 275% IoU improvement and an 82% FID reduction over traditional physics-based inversions, while delivering $256\times256$-resolution outputs and better performance near boundaries and for non-convex shapes. The work demonstrates strong generalization to real-world data and highlights the potential of generative AI to overcome nonlinearity and ill-posedness in PD-ISP for indoor imaging, localization, and tracing. This approach promises privacy-friendly, device-free, high-fidelity indoor imaging with implications for robotics and IoT, and suggests avenues for multi-object, real-time, and 3D extensions.

Abstract

Indoor imaging is a critical task for robotics and internet-of-things. WiFi as an omnipresent signal is a promising candidate for carrying out passive imaging and synchronizing the up-to-date information to all connected devices. This is the first research work to consider WiFi indoor imaging as a multi-modal image generation task that converts the measured WiFi power into a high-resolution indoor image. Our proposed WiFi-GEN network achieves a shape reconstruction accuracy that is 275% of that achieved by physical model-based inversion methods. Additionally, the Frechet Inception Distance score has been significantly reduced by 82%. To examine the effectiveness of models for this task, the first large-scale dataset is released containing 80,000 pairs of WiFi signal and imaging target. Our model absorbs challenges for the model-based methods including the non-linearity, ill-posedness and non-certainty into massive parameters of our generative AI network. The network is also designed to best fit measured WiFi signals and the desired imaging output. For reproducibility, we will release the data and code upon acceptance.

Figures (6)

• Figure 1: DoI with wireless transceiver nodes at its boundary B and dimensions $X$ × $Y$ at cross-sectional height $h$ . The transmitters and receivers are located at the boundary $B$. The whole area $D$ is discretized into pixels by $\Delta x$ and $\Delta y$.
• Figure 2: Framework of our proposed WiFi-GEN, where Downsample, Upsample and Signal Block are illustrated in detail in Figure \ref{['fig:details']}. Both input layer and skip layer consist of simple fully convolutional network.
• Figure 3: Structure details of Downsample, Upsample and Signal Block in WiFi-GEN. $F_i$ denotes the feature extracted by the i-th downsample module, $F_i^{'}$ denotes the feature extracted by the i-th upsample module.
• Figure 4: Performance of selected cases of our model in comparison with physical model-based methods. Our model demonstrates substantial improvement in shape reconstruction accuracy.
• Figure 5: Image generation capability with different noise condition in model training, where w/ referes to with, w/o refers to without