vita-epfl/s-attack

[CVPR 2022] S-attack library. Official implementation of two papers "Vehicle trajectory prediction works, but not everywhere" and "Are socially-aware trajectory prediction models really socially-aware?".

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

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

---

Vehicle trajectory prediction works, but not everywhere, CVPR 2022
M. Bahari, S. Saadatnejad, A. Rahimi, M. Shaverdikondori, A. Shahidzadeh, S. Moosavi-Dezfooli, A. Alahi
Website                 Paper                 Citation                 Code

---

Are socially-aware trajectory prediction models really socially-aware?, TR_C 2022
S. Saadatnejad, M. Bahari, P. Khorsandi, M. Saneian, S. Moosavi-Dezfooli, A. Alahi
Website                 Paper                 Citation                 Code

---

For citation:

@InProceedings{bahari2022sattack,
    author    = {Bahari, Mohammadhossein and Saadatnejad, Saeed and Rahimi, Ahmad and Shaverdikondori, Mohammad and Shahidzadeh, Amir-Hossein and Moosavi-Dezfooli, Seyed-Mohsen and Alahi, Alexandre},
    title     = {Vehicle trajectory prediction works, but not everywhere},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    year      = {2022},
}
@article{saadatnejad2022sattack,
     author = {Saeed Saadatnejad and Mohammadhossein Bahari and Pedram Khorsandi and Mohammad Saneian and Seyed-Mohsen Moosavi-Dezfooli and Alexandre Alahi},
     title = {Are socially-aware trajectory prediction models really socially-aware?},
     journal = {Transportation Research Part C: Emerging Technologies},
     volume = {141},
     pages = {103705},
     year = {2022},
     issn = {0968-090X},
     doi = {https://doi.org/10.1016/j.trc.2022.103705},
}

s-attack's Issues

Extension to argoverse2 dataset

Hi, thank you for the great work.
I want to extend scene attack methodology to ArgoVerse2 and was wondering if provide me with a starting point on what to change and where?
Best,
Ahmed

PECNet and STG-CNN

Dear authors,

I've noticed there is no option in your code to test S-Attack on PECNet and STG-CNN, two methods for which you provide adversarial attack results in your paper. I was wondering if you were planning on making your code for these two available, or if there's some way to install the dependencies separately ourselves.

Thank you,
Erica

How to realise visualization in evaluator files

Hello, your article on pedestrian prediction has inspired me a lot. After debugging the code, I want to find out how to visualize the final result, because only a few data can be seen in the debugging process, and I don't know the implementation process. If you like, can you explain to me in detail how to do the visualization part？

Finetuning model

Is the data that needs to be downloaded in the Finetuning model the data that adds scene enhancement? , what proportion does it account for? What to do if you need to generate it yourself

Permutations on STGCNN and PECNet

Hi, great work! Can you provide the permutation trajectories obtained from PECNet and STGCNN publicly or privately if their code option cannot be released easily? I hope I can use them to do quick collision tests on other models. Thanks.