S-attack library:
A library for evaluating trajectory prediction models

This library contains two research projects to assess the trajectory prediction models, Social-attack which evaluates social understanding of models, and Scene-attack which evaluates the scene-understanding of them.

Are socially-aware trajectory prediction models really socially-aware?
S. Saadatnejad, M. Bahari, P. Khorsandi, M. Saneian, S. Dezfooli, A. Alahi, arxiv 2021
Website                 Paper                 Citation

Vehicle trajectory prediction works, but not everywhere
M. Bahari, S. Saadatnejad, A. Rahimi, M. Shaverdikondori, S. Dezfooli, A. Alahi, arxiv 2021
Website                 Paper                 Citation

Social-attack

Are socially-aware trajectory prediction models really socially-aware?

The official code for the paper: "Are socially-aware trajectory prediction models really socially-aware?", Webpage, arXiv

Installation:

Start by cloning this repository:

git clone https://github.com/vita-epfl/s-attack
cd s-attack

And install the dependencies:

pip install .

For more info on the installation, please refer to Trajnet++

Dataset:

• We used the trajnet++ dataset. For easy usage, we put data in DATA_BLOCK folder.

Training/Testing:

In order to attack the LSTM-based models (S-lstm, S-att, D-pool):

bash lrun.sh

In order to attack the GAN-based models:

bash grun.sh

Scene-attack

Vehicle trajectory prediction works, but not everywhere

The official code for the paper: "Vehicle trajectory prediction works, but not everywhere", Webpage, arXiv

Code will be released soon!

For citation:

@article{saadatnejad2021sattack,
  title={Are socially-aware trajectory prediction models really socially-aware?},
  author={Saadatnejad, Saeed and Bahari, Mohammadhossein and Khorsandi, Pedram and Saneian, Mohammad and Moosavi-Dezfooli, Seyed-Mohsen and Alahi, Alexandre},
  year={2021}, eprint={2108.10879}, archivePrefix={arXiv}, primaryClass={cs.CV}
}
@article{bahari2021sattack,
  title={Vehicle trajectory prediction works, but not everywhere},
  author={Bahari, Mohammadhossein and Saadatnejad, Saeed and Rahimi, Ahmad and Shaverdikondori, Mohammad and Moosavi-Dezfooli, Seyed-Mohsen and Alahi, Alexandre},
  year={2021}, eprint={2112.03909}, archivePrefix={arXiv}, primaryClass={cs.CV}
}