GRaF-pretrained — Pretrained Weights

Pretrained checkpoints for 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  ·  📊 Dataset: kyang73/GRaF-sim

Checkpoints

File	Pretrained on	Intended use
type2_catA.pth	D2–D5 (category A)	Cross-scene generalization within a single room type
type3_AB_testC.pth	D2–D13 (A + B)	Cross-category generalization to a held-out room type
type4_sim2real.pth	D2–D19 (all sims)	Sim-to-real transfer to real RFID measurements (D1)

Training data: kyang73/GRaF-sim.

Download

from huggingface_hub import snapshot_download
snapshot_download("kyang73/GRaF-pretrained", local_dir="pretrained")

Or download a single checkpoint:

from huggingface_hub import hf_hub_download
ckpt = hf_hub_download("kyang73/GRaF-pretrained", "type4_sim2real.pth")

Usage

Point the matching YAML config at the checkpoint via --resume_from:

python inference.py --config_path configs/type4_zeroshot_D1.yaml \
                    --resume_from pretrained/type4_sim2real.pth

Citation

@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

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