RMDirectionalBerlin

A dataset of simulated path loss radio maps for training CNN. The dataset and experiments are described in "Radio Map Estimation - An Open Dataset with Directive Transmitter Antennas and Initial Experiments" (arXiv:2402.00878), the code can be found at https://github.com/fabja19/RML. For updated results on images and applications to coverage optimization, see the updated version of the repo https://github.com/fabja19/RML_v2_img and arxiv:2410.17264.

RMBerlinDirectional.zip:This file contains the dataset used in our experiments, including the radio maps, corresponding city maps, Tx information and aerial images. The dataset class in the code shows how they can be used. Structure:

path_gain - contains the target radio maps as .png images

gis - nDSMs of buildings and vegetation for each map as gray-scale .png images

tx_antennas - .npy files containing Tx locations and characteristics for all samples

img - RGBIR aerial images in .tif format

antenna_patterns  - .npy files containing antenna patterns

los - line-of-sight information for each sample in .npy file

.json files containing lists of the sample ids and corresponding Tx locations and characteristics for the dataset class

model_checkpoints.zip:Contains checkpoints and configs for some trained models.

model_checkpoints_v2_img.zip:Contains updated checkpoints and configs for some trained models for the prediction from images.

additional_gis_files.zip:Contains data that was used to prepare the simulations but that is not needed for running the code.

nDSM_orig - original nDSM files for buildings and vegetation created from LiDAR data, in .tif format

polygons - polygons with height values generated from the original nDSM files and simplified, used in the ray-tracing simulations, in .GeoJSON format