Demo Abstract: AIGC for RFID-based Human Activity Recognition

The lack of sufficient radio frequency (RF) data constrains the performance of AI-empowered wireless communications, networking, and sensing research. RF data collection is more difficult and costly than other data types (e.g., text or image). To this end, we propose to exploit the strength of diffusion models on latent domains to generate super-realistic data for RF sensing applications. In this demo, we present a novel lightweight AIGC framework centered on latent domains, termed Activity Class Conditional Latent Diffusion Model (RFID-ACCLDM), for easy generation of large amounts of RF data at low cost, conditioned on activity class labels. We demonstrate the high performance of RFID-ACCLDM with RFID-based 3D human pose estimation and human activity recognition (HAR) model as representative downstream tasks.