GRaF-sim — Simulated RF Spectrum Dataset

Simulated wireless channel spectra for 18 indoor scenes, released with the CVPR 2026 paper "Generalizable Radio-Frequency Radiance Fields for Spatial Spectrum Synthesis" by Kang Yang, Yuning Chen, and Wan Du.

📄 Paper: arXiv:2502.05708  ·  💻 Code: github.com/kangyang73/GRaF  ·  🤖 Models: kyang73/GRaF-pretrained

Used to train and evaluate GRaF, a generalizable radio field model.

Each scene is modeled as a box room with ITU-material walls and furniture primitives. For every transmitter position, Sionna ray-tracing computes the channel response at an 8 × 8 receiver array, and conventional beamforming over a 360° × 90° azimuth–elevation grid produces the final spatial spectrum image.

Scenes

Each category contains six scene variants with increasing furniture/clutter.

Category	Room	Room size (m)	Scenes
A	Conference Room	8.0 × 6.0 × 3.0	D2–D7
B	Classroom	10.0 × 7.0 × 3.0	D8–D13
C	Bedroom	5.0 × 4.0 × 2.8	D14–D19

Per-scene layout

D<i>/
├── gateway_info.yml   # room size + receiver (gateway) pose
├── tx_pos.csv         # 5 340 transmitter positions (x, y, z)
└── spectrum/
    ├── 00001.png      # 360 × 90 grayscale spectrum (azimuth × elevation)
    ├── 00002.png
    └── ...            # one image per TX position

Totals: 18 scenes × 5 340 samples ≈ 96 K spectrum images, ~785 MB.

Download

from huggingface_hub import snapshot_download
snapshot_download("kyang73/GRaF-sim", repo_type="dataset", local_dir="data")

Or via CLI:

hf download kyang73/GRaF-sim --repo-type dataset --local-dir data

Citation

If you use this dataset, please cite:

@inproceedings{Yang2026_GRaF,
  author    = {Kang Yang and Yuning Chen and Wan Du},
  title     = {Generalizable Radio-Frequency Radiance Fields for Spatial Spectrum Synthesis},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2026},
}

License

Released under the BSD 3-Clause License. Copyright (c) 2025, Kang Yang, Yuning Chen, and Wan Du. See the GRaF repository for full terms.