T-JEPA

This is the official code for T-JEPA: A Joint-Embedding Predictive Architecture for Trajectory Similarity Computation. This paper has been accepted by SIGSPATIAL 2024.

📢 Latest update: The code has been optimized for ~4× faster inference.

If you find this work useful for your research, please cite:

@inproceedings{li2024t,
  title={T-JEPA: A Joint-Embedding Predictive Architecture for Trajectory Similarity Computation},
  author={Li, Lihuan and Xue, Hao and Song, Yang and Salim, Flora},
  booktitle={Proceedings of the 32nd ACM International Conference on Advances in Geographic Information Systems},
  pages={569--572},
  year={2024}
}

Run T-JEPA

To pre-train T-JEPA:

python train.py --dataset porto

And to run downstream fine-tuning (approximate heuristic measures) after pre-training:

python train_trajsimi.py --dataset porto --trajsimi_measure_fn_name hausdorff

We follow the preprocessing protocol of TrajCL. Please refer to TrajCL for data preparation.

Acknowledgement

The code is largely borrowed from TrajCL.