Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Eval5 update carla video attack #1516

Merged

Conversation

yusong-tan
Copy link
Contributor

Add code to initialize the patch to the correct size for Eval5 data

@lcadalzo lcadalzo self-assigned this May 17, 2022
@lcadalzo
Copy link
Contributor

@yusong-tan what effect(s) (qualitative and/or quantitative) did you observe that this change has on the results?

@yusong-tan
Copy link
Contributor Author

@lcadalzo None that I can see on undefended model because it is so brittle against perturbation, even ones that are significantly distorted due to resizing. In this fix, the difference between initial patch size and target embedded size will be minimized, so we should be able to produce stronger attacks against defended models.

@lcadalzo
Copy link
Contributor

I'm observing 17.05% adversarial_video_tracking_mean_iou on the undefended model, slightly below the current baseline of 18%

@lcadalzo lcadalzo merged commit b82af26 into twosixlabs:develop May 18, 2022
@yusong-tan yusong-tan deleted the eval5_update_carla_video_attack branch January 12, 2023 21:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

2 participants