Streaming Real-Time Trajectory Prediction Using Endpoint-Aware Modeling (SEAM)

(Code and model weights coming soon!)

Paper

Streaming Real-Time Trajectory Prediction Using Endpoint-Aware Modeling
Alexander Prutsch, David Schinagl, Horst Possegger Graz University of Technology
WACV 2026 Oral

Getting Started

Create and Activate Virtual Environment

conda create -n seam python=3.11.10
conda activate seam

Install PyTorch

We tested our implementation with torch 2.1.1 and CUDA 12.1.

Install PyTorch e.g.

conda install pytorch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 pytorch-cuda=12.1 -c pytorch -c nvidia

Install Dependencies

pip install -r ./requirements.txt

Download Download Argoverse 2 Motion Forecasting Dataset

The expected structure of the AV2 data should be:

data_root
    ├── train
    │   ├── 0000b0f9-99f9-4a1f-a231-5be9e4c523f7
    │   ├── 0000b6ab-e100-4f6b-aee8-b520b57c0530
    │   ├── ...
    ├── val
    │   ├── 00010486-9a07-48ae-b493-cf4545855937
    │   ├── 00062a32-8d6d-4449-9948-6fedac67bfcd
    │   ├── ...
    ├── test
    │   ├── 0000b329-f890-4c2b-93f2-7e2413d4ca5b
    │   ├── 0008c251-e9b0-4708-b762-b15cb6effc27
    │   ├── ...

Data Preprocessing

Preprocess the Argoverse 2 dataset by executing

python preprocess.py --data_root=/path/to/data_root -p

Training on Single-Agent benchmark

Train SEAM model using

python train.py datamodule.pl_module.data_root=/path/to/data_root/seam_processed/

Evaluation on Single-Agent benchmark

Evaluate SEAM model using

python train.py datamodule.pl_module.data_root=/path/to/data_root/seam_processed/ checkpoint=outputs/path/to/experiment/checkpoint_file.ckpt

Visualize Results

Visualize the prediction results using

python visualize.py

Please update the data_root, chkpt_dir, and av2_raw_data_dir variable in the script.

Bibtex

@inproceedings{prutsch2026streaming,
 title={{Streaming Real-Time Trajectory Prediction Using Endpoint-Aware Modeling}},
 author={Alexander Prutsch, David Schinagl, Horst Possegger},
 booktitle={WACV},
 year={2026},
}

Acknowledgements

This repository is based on RealMotion and integrates code from Forecast-MAE and EMP. We thank them for their work!