Research map / 3D vision / Gaussian splatting
RxGS: Receiver-Generalizable 3D Gaussian Splatting for Radio-Frequency Data Synthesis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes RxGS, a two-stage 3DGS framework that learns receiver-independent Gaussian geometry and then receiver-conditioned radiance (global and local modulation), enabling one shared model for many receivers and unseen-receiver synthesis for RF data.
Problem
Existing 3DGS-based RF data synthesis methods model only the transmitter dependence, so each receiver needs its own model and unseen receiver positions cannot be predicted.
Approach
Split Gaussian attributes into receiver-independent geometry (learned in Stage I from one reference receiver, then frozen) and receiver-dependent radiance (Stage II), with a global affine modulation branch and a local per-Gaussian branch driven by direction, distance and occlusion features, plus a multi-receiver CUDA rasterizer that shares projection and sorting across receivers.
Key results
- On BLE RSSI, RxGS on GSRF reduces MAE from 3.74 to 3.57 dBm with a single shared model versus per-receiver GSRF.
- On unseen receivers, RxGS achieves 4.92 dBm MAE on BLE versus 9.7-11.6 dBm for per-receiver 3DGS baselines.
- Training cost reduced by up to 45x, inference by 7.6x, and storage by roughly N times relative to per-receiver baselines on RFID with N=21.
Surprising findings
- RxGS on RF-3DGS and WRF-GS+ slightly degrades RFID spectrum and WiFi CSI quality relative to per-receiver baselines
- Joint training of geometry and radiance (no two-stage) substantially worsens BLE MAE (4.44 vs 3.57 dBm)
Limitations
- Assumes a static scene; cannot handle moving people or rearranged furniture
- Conditioning generalizes within a scene but not across scenes, requiring a site survey per environment
- Trails per-receiver baselines by 0.8 to 1.7 dB on spectrum and CSI datasets
Experimental setup
Datasets: BLE RSSI dataset RFID spatial spectrum simulated with Sionna WiFi CSI dataset
Benchmarks: BLE RSSI RFID spatial spectrum WiFi CSI
Metrics: MAE MSE PSNR SSIM SNR training time inference time storage localization error
Baselines: NeRF^2 GRaF RF-3DGS GSRF WRF-GS+
Builds on: 3D Gaussian Splatting RF-3DGS GSRF WRF-GS+
Closest papers by meaning
- pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction (CVPR 2024 · similarity 0.90)
- GSRF: Complex-Valued 3D Gaussian Splatting for Efficient Radio-Frequency Data Synthesis (NeurIPS 2025 · similarity 0.89)
- COLMAP-Free 3D Gaussian Splatting (CVPR 2024 · similarity 0.89)
- Gaussian Herding across Pens: An Optimal Transport Perspective on Global Gaussian Reduction for 3DGS (NeurIPS 2025 · similarity 0.89)
- Structure Consistent Gaussian Splatting with Matching Prior for Few-shot Novel View Synthesis (NeurIPS 2024 · similarity 0.89)
- GaussianPro: 3D Gaussian Splatting with Progressive Propagation (ICML 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.