Yasoz / DiffTraj

The implemention of DiffTraj: Generating GPS Trajectory with Diffusion Probabilistic Model
MIT License
51 stars 10 forks source link

Requirements

You can install the required packages in your python environment by:

  python>=3.7
  torch>=1.7
  pandas
  numpy
  matplotlib
  pathlib
  shutil
  datetime
  colored
  math

Train the trajectory generation model

  1. Download the data with permission of using your own GPS trajectory data

  2. Change the corresponding paths in the config.py and main.py files

  3. Run and train the model by:

    python main.py 
  4. The code will save the trained model every 10 epoches

Usage for Trajectory Generation

Refer to Traj_Generation.ipynb file, we provide the fine-trained model (model.pt) simple guide infromation(heads.npy) example to explain how to generate trajectories.

Generated result

Chengdu

Reference

If you use this code, please cite:

@inproceedings{zhu2023DiffTraj,
    author = {Yuanshao Zhu, Yongchao Ye, Shiyao Zhang, Xiangyu Zhao and James, J.Q. Yu},
    title = {DiffTraj: Generating GPS Trajectory with Diffusion Probabilistic Model},
    booktitle = {Proceedings of the 37th Annual Conference on Neural Information Processing Systems},
    year = {2023}
}