• This is the official implementation of the paper: S2TNet: Spatio-Temporal Transformer Networks for Trajectory Prediction in Autonomous Driving (ACML 2021).

Requires:

• adamod==0.0.3
• ConfigArgParse==1.5.2
• numpy==1.19.0
• PyYAML==6.0
• scipy==1.7.1
• tensorboardX==2.5.1
• torch==1.9.0
• tqdm==4.31.1

 pip install -r requirements.txt

We use Apollo Scape Trajectory dataset

Results on Apollo Scape:

You can train our model by below command:

python3 main.py --config ./config/apolloscape/train.yaml

You can test our model by below command:

python3 main.py --config ./config/apolloscape/test.yaml

The result file, named as prediction_result.zip, is generated after testing phase. Then, you can directly upload the file to (http://apolloscape.auto/trajectory.html) to obtain the official results.

If you find our work useful for your research, please consider citing the paper:

@inproceedings{pmlr-v157-chen21a,
  title = 	 {S2TNet: Spatio-Temporal Transformer Networks for Trajectory Prediction in Autonomous Driving},
  author =       {Chen, Weihuang and Wang, Fangfang and Sun, Hongbin},
  booktitle = 	 {Proceedings of The 13th Asian Conference on Machine Learning},
  pages = 	 {454--469},
  year = 	 {2021},
  volume = 	 {157},
  month = 	 {17--19 Nov}
}