Radio map construction via generative diffusion models — UNIC Lab, Xidian University

RadioDiff — The foundational diffusion model for radio map construction.   📄 Paper  |  💻 Code  | 

RadioDiff-k² — PINN-enhanced diffusion guided by the Helmholtz equation.   📄 Paper  |  💻 Code  | 

iRadioDiff — Indoor radio map construction with physical information integration.   📄 Paper  |  💻 Code  |   

RadioDiff-Turbo — Efficiency-enhanced RadioDiff for accelerated inference.   📄 Paper  | 

RadioDiff-Flux — Adaptive reconstruction under dynamic environments and base station location changes.   📄 Paper  | 

RadioDiff-3D — 3D radio map construction with the UrbanRadio3D dataset.   📄 Paper  |  💻 Code  | 

RadioDiff-FS — Few-shot learning for radio map construction with limited measurements.   📄 Paper  |  💻 Code  | 

RadioDiff-Inverse — Sparse measurement-based radio map recovery for ISAC applications.   📄 Paper  |  💻 Code  | 

RadioDiff-Loc — Sparse measurement-based NLoS localization using diffusion models.   📄 Paper  | 

📚 For a comprehensive categorized overview of radio map research, visit Awesome-Radio-Map-Categorized.

Diffusion posterior sampling for radio-map style inpainting / inverse problems.

This repository applies the FPS-SMC diffusion posterior sampling pipeline to radio map reconstruction tasks, using building masks as conditioning signals and the RadioMapSeer dataset.

conda create -n radiodiff python=3.8 -y
conda activate radiodiff
pip install -r requirements.txt

A CUDA-compatible PyTorch build is strongly recommended. The code was tested with PyTorch 2.0.0+cu118.

This project uses the RadioMapSeer dataset for radio map reconstruction experiments.

• Homepage: https://radiomapseer.github.io/
• Download: Follow the instructions on the dataset homepage to obtain the data.

After downloading, organize the data under data/ as follows:

data/
├── samples/           # radio map images (used as ground truth)
├── buildings_complete/# building footprint masks
├── antennas/          # antenna position maps
└── val_images/        # validation split images

See data/README.md for more details.

Download the pretrained score estimation models and place them in models/.

models/
├── ffhq_10m.pt
└── imagenet256.pt      # or 256x256_diffusion_uncond.pt

See models/README.md for details.

All scripts default to repository-relative output paths under results/. You can override the output directory by passing it as the first argument.

./run_random_cond.sh
# or specify output directory:
./run_random_cond.sh results/my_run

./run_random_uncond.sh

./run_sensor_cond.sh

./run_sensor_uncond.sh

python3 sample_condition.py \
    --task_config configs/inpainting_config_random_cond.yaml \
    --save_dir results/my_run \
    --gpu 0 \
    --num_images 10 \
    --mask_ratios 0.9 \
    --noise_levels 0.05

python3 sample_interpolation.py \
    --task_config configs/inpainting_config_random_cond.yaml \
    --save_dir results/interp \
    --gpu 0

Evaluate all experiments under results/:

python3 advanced_metric_calculator.py --base_dir results --gpu 0

Evaluate a single experiment directory:

python3 advanced_metric_calculator.py --exp_dir results/random_cond --gpu 0

Outputs per experiment: metrics.json, per_sample_metrics.json Summary outputs: all_result.json, summary_metrics_vMMDD_HHMM.json

.
├── sample_condition.py        # main conditional pipeline
├── sample_condition_uncond.py # main unconditional pipeline
├── sample_interpolation.py    # interpolation baselines (RBF, Linear, etc.)
├── advanced_metric_calculator.py
├── configs/                   # task / model / diffusion YAML configs
├── data/                      # datasets (git-ignored except README)
├── models/                    # checkpoints (git-ignored except README)
├── results/                   # experiment outputs (git-ignored except README)
├── guided_diffusion/          # diffusion model and sampler internals
├── util/                      # metrics, image utilities, logging
├── ImageNet/                  # ImageNet-specific entry scripts
├── scripts/                   # utility scripts
├── run_random_cond.sh
├── run_random_uncond.sh
├── run_sensor_cond.sh
└── run_sensor_uncond.sh

The road-conditioned path is kept for reference but is not part of the default reproducible pipeline:

• run_cond_road.sh is intentionally disabled
• sample_road.py requires additional dataset wiring before use

This codebase builds on the diffusion posterior sampling / FPS-SMC framework. Pretrained checkpoints are from DPS2022. Radio map data is from the RadioMapSeer dataset.