Multimodal Dataset for Sparse Radio Environment Map Reconstruction

This dataset provides a multimodal benchmark for reconstructing high-resolution Radio Environment Maps (REMs) from sparse wireless signal measurements. It consists of spatially aligned triplets of (1) simulated Received Signal Received Power (RSRP) maps generated via electromagnetic modeling, (2) Sentinel-1 Synthetic Aperture Radar (SAR) imagery acquired from Google Earth Engine, and (3) rasterized building information maps derived from OpenStreetMap. Each sample is standardized to 256\u00d7256 resolution and aligned to an urban layout, facilitating deep learning-based REM estimation under extreme sampling sparsity. The dataset is designed to support research in multimodal fusion, wireless propagation modeling, and IoT-based signal environment mapping. It serves as the experimental foundation for the CMCTNet model proposed in our accompanying manuscript. This resource is ideal for evaluating and benchmarking algorithms in signal recovery, environmental awareness, and intelligent wireless system design.