DAM4SAM: Distractor-aware memory for SAM2

DAM4SAM (CVPR25) implementation that supports tracking multiple objects simultaneously. The original implementation of the method is available in this repository.

If you find this work interesting, please cite the original publication:

@InProceedings{dam4sam,
  author = {Videnovic, Jovana and Lukezic, Alan and Kristan, Matej},
  title = {A Distractor-Aware Memory for Visual Object Tracking with {SAM2}},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2025},
  pages = {24255-24264}
}

Instructions

Run the following command to test it on a single VOTS2025/2024/2023 sequence:

CUDA_VISIBLE_DEVICES=0 python run_on_vot_multi.py --dataset <dataset_dir> --visualize --sequence <sequence_name>

Optionally, you can set also the --checkpoint_dir input argument passing the path to the directory containing checkpoints.

VOTS 2025 integration

Use the vot_wrapper.py script to integrate it with the vot toolkit and running experiments. More information about the VOTS challenges is available on the official VOTS webpage.

Results

VOTS2025 baseline experiment:

Model	Q	Acc.	Rob.
SAM2.1-L	0.710	0.791	0.862
SAM2.1-T	0.666	0.759	0.847

VOTS2025 realtime experiment (note that the realtime results might vary depending on the hardware):

Model	Q	Acc.	Rob.
SAM2.1-L	0.626	0.687	0.862
SAM2.1-T	0.632	0.714	0.852