Skip to content

[NeurIPS 2024 in AdvML Workshop] TrackPGD: Efficient Adversarial Attack using Object Binary Masks against Robust Transformer Trackers

License

Notifications You must be signed in to change notification settings

fatemehN/TrackPGD

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 

Repository files navigation

TrackPGD Attack

[NeurIPS 2024 in AdvML Workshop] TrackPGD: Efficient Adversarial Attack using Object Binary Masks against Robust Transformer Trackers

Project Webpage

OpenReview

ArXiv

Mask Evaluation

Step 1: Download the trackers packages

Please download the trackers from the VOT challenge (VOT2022) website, as follows:

Step 2: Create the environment

For each tracker follow the instructions to build the suitable environment as stated in their README.md file.

Step 3: Download the networks

For TrackPGD experiments, we used the following networks:

Step 4: Run the setup files

Follow the instructions of each tracker to correct the paths and run the setup files.

Step 5.a: Set the paths for MixFormerM

1- From MixFormerM folder on TrackPGD directory(TrackPGD/MixFormerM), find TrackPGD folder. Copy and paste this folder into the tracker folder (/MixFormerM_submit/mixformer/external/AR/pytracking/).

2- Add a new entry to the trackers.ini file in the "vot22_seg_mixformer_large" directory(/MixFormerM_submit/mixformer/vot22_seg_mixformer_large) as follows:

[MixFormer_TrackPGD]  
label = MixFormer_TrackPGD
protocol = traxpython
command = mixformer_vit_large_vit_seg_class_TrackPGD
paths = <PATH_OF_MIXFORMER>:<PATH_OF_MIXFORMER>/external/AR/pytracking/TrackPGD:<PATH_OF_MIXFORMER>/external/AR
env_PATH = <PATH_OF_PYTHON>

3- Edit the paths of MixFormer_TrackPGD entry to include all of the necessary paths as recommended on the tracker' README.md file. The <PATH_OF_MIXFORMER> is your path to "mixformer" folder.

4- Edit <PATH_OF_PYTHON> with your path to the MixFormer environment you built in ##Step 2.

Step 5.b: Set the paths for OSTrackSTS

1- From OSTrackSTS folder on TrackPGD directory(TrackPGD/OSTrackSTS), find TrackPGD folder. Copy and paste this folder to the (OSTrack/external/AR_VOT22/pytracking) of the tracker folder.

2- Add a new entry to the trackers.ini file in the "vot22/OSTrackSTS" directory(OStrack/external/vot22/OSTrackSTS) as follows:

[OSTrackSTS_TrackPGD]  
label = OSTrackSTS_TrackPGD
protocol = traxpython
command = OSTrackSTS_TrackPGD
paths = <PATH_OF_OSTrack>:<PATH_OF_OSTrack>//external/AR_VOT22/pytracking/TrackPGD
env_PATH = <PATH_OF_PYTHON>

3- Edit the paths of the OSTrackSTS_TrackPGD entry to include all of the necessary paths as recommended on the tracker' README.md file. The <PATH_OF_OSTrack> is your path to the "OSTrack" folder.

4- Edit <PATH_OF_PYTHON> with your path to the OSTrack environment you built in ##Step 2.

Step 6.a: Run the MixFormerM tracker attacked by TrackPGD for VOT2022STS evaluation

1- Enter the VOT workplace directory (/path/to/vot22_seg_mixformer_large)

2- Activate the MixFormer environment.

3- Run:

vot evaluate --workspace . MixFormer_TrackPGD
vot analysis --workspace .

Step 6.b: Run the OSTrackSTS tracker attacked by TrackPGD for VOT2022STS evaluation

1- Enter the VOT workspace directory (/path/to/vot22/OSTrackSTS/) 2- Activate the OSTrack environment. 3- Run:

vot evaluate --workspace . OSTrackSTS_TrackPGD
vot analysis --workspace .

Contact:

Fatemeh Nokabadi

About

[NeurIPS 2024 in AdvML Workshop] TrackPGD: Efficient Adversarial Attack using Object Binary Masks against Robust Transformer Trackers

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages