PathFinder: Advancing Path Loss Prediction for Single-to-Multi-Transmitter Scenario

This repository contains the official implementation of the paper "PathFinder: Advancing Path Loss Prediction for Single-to-Multi-Transmitter Scenario".

Project page: https://emorzz1g.github.io/PathFinder

Dynamic Demonstrations

Real-time path loss prediction with a walking transmitter across different urban layouts.

Highlights

• Multi-GPU training and inference via PyTorch DDP.
• Config-driven experiments (YAML).
• Baselines and ablations supported by swapping config or entry scripts.

Requirements

• Python 3.8+ (tested with PyTorch)
• CUDA-capable GPU(s) for training

Install dependencies:

pip install -r requirements.txt

Project Structure

• config/: YAML experiment configs
• models/: model code
• logs/: training logs (created at runtime)
• results/: inference outputs (created at runtime)
• trainer/: training and inference logic

Pre-trained Model Link

You can download our pre-trianed models from Google Driver.

https://drive.google.com/drive/folders/1CBaQWyIV5sb2xLmvvwKS_QKQBOZ7ZEov?usp=sharing

Data

Set the dataset path in your config file (see dataset.dataset_dir in config/default_pathfinder.yaml).

The method for downloading the dataset will be sorted out later.

Quick Start

Train PathFinder:

python main.py --config default_pathfinder.yaml --mode train

Test / Inference:

python main.py --config default_pathfinder.yaml --mode test

Baselines

Baseline configurations are provided under config/. To run a baseline, switch the config file. Example:

python main.py --config default_radiounet.yaml --mode train

The baseline_main.py entry point uses a different trainer implementation and can be used for legacy runs.

Configuration

Key options in YAML:

• model.name: model type (e.g., PathFinder, UNet, RadioUNet, PMNet, REM_Net)
• opti: optimizer settings and training schedule
• dataset: dataset path and split indices
• load_pretrain and ckp_name: control checkpoint loading

Outputs

• Checkpoints: model_path in config (default models/)
• Logs: log_path in config (default logs_pf/ or logs_rem/)
• Results: save_path in config (default results/)

Citation

If you use this code, please cite the paper:

@article{zhong2025pathfinder,
  title={PathFinder: Advancing Path Loss Prediction for Single-to-Multi-Transmitter Scenario},
  author={Zhong, Zhijie and Yu, Zhiwen and Li, Pengyu and Lv, Jianming and Chen, CL and Chen, Min},
  journal={arXiv preprint arXiv:2512.14150},
  year={2025}
}

License

See the LICENSE file for details.