RadCloudSplat

Official implementation of:

RadCloudSplat: Scatterer-Driven 3D Gaussian Splatting with Point-Cloud Priors for Radiomap Extrapolation

Yiheng Wang, Ye Xue, Shutao Zhang, Hongmiao Fan and Tsung-Hui Chang

Thanks for your interest in our work. This repository contains code and links to the RadCloudSplat method for radiomap extrapolation, which has been accepted by IEEE INFOCOM 2026.

More information can be refered to:

• Paper: arXiv
• Published version: IEEE Xplore
• Code: This repository
• Project page: Project Page
• License: MIT

Introduction

In this work, we first extended 3DGS to the radio frequency domain, leveraging camera-free RadCloudSplat to extrapolate RSSs with high accuracy from sparse measurements in an outdoor environment. By efficiently selecting the means of key virtual scatterers from dense point clouds aided by the relaxed-mean (RM) scheme, the model captured intricate multi-path propagation characteristics. Experiments and analysis validated the effectiveness of these scatterers, advancing the state-of-the-art in wireless network modeling and extrapolation performance and highlighting the transformative potential of integrating advanced 3D modeling techniques with wireless propagation analysis for next-generation applications in the radio domain.

Schematic illustration of RadCloudSplat, comprising three major parts: 1) Relaxed-Mean Reparameterization for Key Virtual Scatters Positions Extraction. 2) Camera-Free RadCloudSplat Model for RSS Synthesis. 3) Optimizing RadCloudSplat Scheme

News

Due to copyright issues regarding the measurement real data from wireless network, we are unable to provide the data used in the paper.

• Release the demo dataset revised from NeRF2 datasets.
• Release the training code.
• Release the inference code.
• Release the paper of RadCloudSplat on arXiv.

Training & Evaluation

A small demo dataset in ./demo_data is included to help quickly verify the code, which can be executed using the following command:

python train_radcloudsplat.py

More datasets can be found here.

When you finish the training, you can inference the trained model by using the following command:

python train_radcloudsplat.py --mode test

Citation

If you find our work useful in your research, please consider citing RadCloudSplat:

@inproceedings{wang2026radcloudsplat,
  title={RadCloudSplat: Scatterer-Driven 3D Gaussian Splatting with Point-Cloud Priors for Radiomap Extrapolation},
  author={Wang, Yiheng and Xue, Ye and Zhang, Shutao and Fan, Hongmiao and Chang, Tsung-Hui},
  booktitle={IEEE INFOCOM 2026-IEEE Conference on Computer Communications},
  pages={1--10},
  year={2026},
  organization={IEEE}
}

Acknowledgement

We thank Dr. Xiaopeng Zhao, the authors of NeRF2, for making their code and dataset available.