Set up the environment

method 1: using conda

conda env create -f environment.yaml
conda activate tame

method 2: using pip

conda create -n tame python=3.10
conda activate tame
pip install -r requirements.txt

Prepare the dataset

Download the NGSIM US-101 and I-80 dataset from here and highD dataset from here. Then use preprocess_*.m in the data folder to preprocess the dataset.

data structure

data
├── ngsimdata
│   ├── TrainSet.mat
│   ├── TestSet.mat
│   ├── ValSet.mat
├── highDdata
│   ├── TrainSet.mat
│   ├── TestSet.mat
│   ├── ValSet.mat

Train the model

cd method
bash train.sh # train the model in DDP mode
# or
python train.py # train the model in single GPU mode

Evaluate the model

cd method
python evaluate.py 

Citation

If you find this work useful, please consider citing (coming soon):

@article{jiang2024hybrid,
  title={Hybrid Attention-based Multi-task Vehicle Motion Prediction Using Non-Autoregressive Transformer and Mixture of Experts},
  author={Jiang, Hao and Hu, Chuan and Niu, Yixun and Yang, Biao and Chen, Hao and Zhang, Xi},
  journal={IEEE Transactions on Intelligent Vehicles},
  year={2024},
  publisher={IEEE}
}

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.