[NeurIPS 2024 in AdvML Workshop] Adversarial Bounding Boxes Generation (ABBG) Attack against Visual Object Trackers
Please download the trackers from the VOT challenge (VOT2022) website and GitHub, as follows:
1- ROMTrack: https://github.com/dawnyc/ROMTrack/archive/refs/heads/master.zip
2- TransTM: https://data.votchallenge.net/vot2022/trackers/TransT_M-code-2022-05-03T05_05_59.924051.zip
For each tracker follow the instructions to build the suitable environment as stated in their README.md file.
For ABBG experiments, we used the following networks:
-
ROMTrack:
-
Tracker network(ROMTrack-Base/ROMTrack/GOT-10k/ROMTrack_epoch0050.pth.tar) from https://drive.google.com/drive/folders/1Q7CpNIhWX05VU7gECnhePu3dKzTV_VoK?usp=drive_link
Download the network and place the file into the following directory: "ROMTrack/lib/models"
-
-
TransTM:
- Tracker network is on the directory of (TransT_M/models/TransTiouhsegm_ep0090.pth.tar)
Follow the instructions of each tracker to correct the paths and run the setup files.
1- Copy the ABBG folder from the ABBG/ROMTrack directory and paste it into the tracker folder at ROMTrack/lib/test/tracker.
2- Copy the tracker_ABBG.py from the ABBG/ROMTrack/ directory and paste it into the evaluation folder at ROMTrack/lib/test/evaluation.
3- Copy the test_ABBG.py from the ABBG/ROMTrack/ directory and paste it into the tracking folder located at ROMTrack/tracking
1- Copy the ABBG folder from the ABBG/TransTM directory and paste it into the trackers folder located at TransT_M/pysot_toolkit/trackers.
2- Copy the test_ABBG.py from the ABBG/TransTM directory and paste it into the pysot_toolkit folder located at TransT_M/pysot_toolkit.
3- Set the paths for the dataset (dataset_root) in line 46 and model (net_path) in line 53 of the test_ABBG.py file.
1- Download the GOT10k dataset (http://got-10k.aitestunion.com/downloads) and set the dataset paths in ROMTrack/lib/test/evaluation/local.py.
2- Activate the ROMTrack environment.
3- From the ROMTrack directory, run:
python tracking/test_ABBG.py ROMTrack got_stage2 --dataset got10k_val --params__model ROMTrack_epoch0050.pth.tar
1- Download the GOT10k dataset (http://got-10k.aitestunion.com/downloads) and set the dataset path in TransT_M/pysot_toolkit/test_ABBG.py.
2- Activate the TransTM environment.
3- From the TransT_M directory, run:
python pysot_toolkit/test_ABBG.py --dataset GOT-10k