Generalizable Indoor Path Loss Prediction

This paper is presented in the context of the First Indoor Pathloss Radio Map Prediction Challenge at IEEE ICASSP 2025. We propose a deep learning approach using a customized ResUNet architecture with physics-informed features for predicting radio maps in indoor environments. Our architecture progressively adapts to handle increasing task complexity, incorporating dual-stream processing for frequency generalization and specialized antenna gain processing with dilated convolutions. Experimental results demonstrate effective generalization across unknown indoor scenes, frequencies, and antenna patterns while maintaining computational efficiency.