"When Einstein Meets Deep Learning" — We make radio signals elegantly dance to the laws of physics in virtual cities. 💃

🎉 [NEW!] Our paper has been accepted by ACM MM BNI 2025 as an Oral Presentation! 🏆
🔓 All code is now available — Ready for researchers and practitioners to explore and build upon our work! 💻
🎯 Pre-trained model weights are now available — Download from Baidu Netdisk: Link (Code: dnd4) 📦

• 🧠 Physics-Informed AI: Equipping neural networks with electromagnetic wisdom, enabling AI to think using Helmholtz equations.
• 🎭 Dual U-Net Architecture: Two neural nets—one handling physical laws, the other refining details—working seamlessly to reconstruct radio maps.
• 📉 Record-Breaking Accuracy: Achieved an unprecedented 0.0031 NMSE error in static scenarios, 2× better than state-of-the-art methods!
• 🌪️ Dynamic Scene Mastery: Robust reconstruction in dynamic, interference-rich environments (vehicles, moving obstacles) with an impressive 0.0047 NMSE.
• 🕵️ Sparse Data Champion: Capable of accurately reconstructing complete radio maps even from a mere 1% sampling—like Sherlock deducing from minimal clues.

• 🧩 Signal Reconstruction Puzzle: Restoring complete electromagnetic fields from fragmented measurements.
• 🌆 Urban Maze Complexity: Seamlessly handling complex obstructions from buildings, moving vehicles, and urban environments.
• ⚡ Real-Time Performance: Achieving inference speeds up to 10× faster than traditional methods—ideal for real-time 5G/6G applications.

1. Physics-Conductor U-Net: Embeds physical laws (Helmholtz equations) through Physics-Informed Neural Networks (PINNs).
2. Detail-Sculptor U-Net: Uses advanced diffusion models for ultra-fine precision in radio map reconstruction.

• 🎯 Anchor Conditional Mechanism: Precisely locking onto critical physical landmarks (like GPS for radio signals).
• 🌐 RF-Space Attention: Models "frequency symphonies" enabling focused learning of electromagnetic signal characteristics.
• ⚖️ Multi-Objective Loss: Harmonizing physics-based constraints and data-driven fitting to achieve optimal results.

Leveraged the authoritative RadioMapSeer dataset:

• 700+ real-world urban scenarios (London, Berlin, Tel Aviv, etc.)
• 80 base stations per map with high-resolution 256×256 grids
• Incorporates static and dynamic challenges (buildings, vehicles)

• Python: 3.8+
• PyTorch: 2.0+ with CUDA support
• GPU: NVIDIA GPU with 8GB+ VRAM (recommended)
• Dataset: RadioMapSeer dataset

1. Clone the repository:

git clone git@github.com:Hxxxz0/RMDM.git
cd PhyRMDM

2. Create and activate conda environment:

conda create -n RMDM python=3.8
conda activate RMDM

3. Install dependencies:

# Install PyTorch with CUDA support
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

# Install all other dependencies
pip install -r requirement.txt

4. Verify installation:

python -c "import torch; print(f'PyTorch {torch.__version__}, CUDA: {torch.cuda.is_available()}')"

Download and organize the RadioMapSeer dataset:

RadioMapSeer/
├── gain/
│   ├── DPM/
│   ├── IRT2/
│   └── cars*/
├── png/
│   ├── buildings_complete/
│   ├── antennas/
│   └── cars/
└── dataset.csv

Single GPU Training (SRM):

conda activate RMDM
python train.py \
  --data_name Radio \
  --data_dir /path/to/RadioMapSeer \
  --batch_size 16 \
  --mixed_precision no \
  --use_checkpoint True \
  --num_channels 96 \
  --attention_resolutions 16 \
  --log_interval 50 \
  --max_steps 100000 \
  --save_interval 10000 \
  --save_dir ./checkpoints_phy

Multi-GPU Training (SRM):

accelerate launch --num_processes=2 --multi_gpu --mixed_precision=no \
  train.py \
  --data_name Radio \
  --data_dir /path/to/RadioMapSeer \
  --batch_size 32 \
  --mixed_precision no \
  --use_checkpoint True \
  --num_channels 96 \
  --attention_resolutions 16 \
  --log_interval 50 \
  --max_steps 100000 \
  --save_interval 10000 \
  --save_dir ./checkpoints_phy

Resume Training:

python train.py \
  --resume_from ./checkpoints_phy/model_phy_step5000.pth \
  --data_name Radio \
  --data_dir /path/to/RadioMapSeer \
  --batch_size 16 \
  --mixed_precision no \
  --save_dir ./checkpoints_phy

Quick Inference Test (SRM):

python sample_test.py \
  --scheduler_type ddpm \
  --data_dir /path/to/RadioMapSeer \
  --checkpoint_path ./checkpoints_phy/model_phy_step100000.pth \
  --output_dir ./inference_results \
  --ddpm_steps 1000 \
  --batch_size 4 \
  --num_samples 100

Full Test Set Evaluation (SRM):

python sample_test.py \
  --scheduler_type ddpm \
  --data_dir /path/to/RadioMapSeer \
  --checkpoint_path ./checkpoints_phy/model_phy_step10000.pth \
  --output_dir ./full_evaluation \
  --ddpm_steps 1000 \
  --batch_size 4 \
  --num_samples -1

Inference with Image Saving 🖼️:

python sample_test.py \
  --scheduler_type ddpm \
  --data_dir /path/to/RadioMapSeer \
  --checkpoint_path ./checkpoints_phy/model_phy_step10000.pth \
  --output_dir ./results_with_images \
  --ddpm_steps 1000 \
  --batch_size 4 \
  --num_samples 50 \
  --save_images

This will generate:

• 📊 Generated radio maps: generated/ folder
• 🎯 Ground truth maps: ground_truth/ folder
• 🏗️ Input conditions: conditions/ folder (buildings + transmitters)
• 🔍 Comparison plots: comparison/ folder (generated vs. ground truth vs. difference)

@misc{jia2025rmdmradiomapdiffusion,
      title={RMDM: Radio Map Diffusion Model with Physics Informed}, 
      author={Haozhe Jia and Wenshuo Chen and Zhihui Huang and Hongru Xiao and Nanqian Jia and Keming Wu and Songning Lai and Yutao Yue},
      year={2025},
      eprint={2501.19160},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2501.19160}, 
}

Special thanks to:

• 🏫 Joint Laboratory of Hong Kong University of Science and Technology (Guangzhou) & Shandong University
• 🌉 Guangzhou Education Bureau's Key Research Project
• 🤖 DIILab for generous computational support

License: This project is distributed under the Academic Free License v3.0. Please cite accordingly for academic use. For commercial applications, contact the authors directly.