Home

Awesome

Bi-directional Adapter for Multimodal Tracking

The official implementation for the AAAI2024 paper Bi-directional Adapter for Multimodal Tracking.

Models

Models & Raw Results (Baidu Driver: ak66)

Models & Raw Results (Google Drive)

Usage

Installation

Create and activate a conda environment:

conda create -n bat python=3.7
conda activate bat

Install the required packages:

bash install_bat.sh

Data Preparation

Download the training datasets, It should look like:

$<PATH_of_Datasets>
    -- LasHeR/TrainingSet
        |-- 1boygo
        |-- 1handsth
        ...
    -- VisEvent/train
        |-- 00142_tank_outdoor2
        |-- 00143_tank_outdoor2
        ...
        |-- trainlist.txt

Path Setting

Run the following command to set paths:

cd <PATH_of_BAT>
python tracking/create_default_local_file.py --workspace_dir . --data_dir <PATH_of_Datasets> --save_dir ./output

You can also modify paths by these two files:

./lib/train/admin/local.py  # paths for training
./lib/test/evaluation/local.py  # paths for testing

Training

Dowmload the pretrained foundation model (OSTrack) (Baidu Driver: 4lvo) / foundation model (Google Drive) and put it under ./pretrained/.

bash train_bat.sh

You can train models with various modalities and variants by modifying train_bat.sh.

Testing

For RGB-T benchmarks

[LasHeR & RGBT234]
Modify the <DATASET_PATH> and <SAVE_PATH> in./RGBT_workspace/test_rgbt_mgpus.py, then run:

bash eval_rgbt.sh

We refer you to use LasHeR Toolkit for LasHeR evaluation, and refer you to use MPR_MSR_Evaluation for RGBT234 evaluation.

For RGB-E benchmark

[VisEvent]
Modify the <DATASET_PATH> and <SAVE_PATH> in./RGBE_workspace/test_rgbe_mgpus.py, then run:

bash eval_rgbe.sh

We refer you to use VisEvent_SOT_Benchmark for evaluation.

Citation

Please cite our work if you think it is useful for your research.

@inproceedings{BAT,
  title={Bi-directional Adapter for Multimodal Tracking},
  author={Bing Cao, Junliang Guo, Pengfei Zhu, Qinghua Hu},
  booktitle={AAAI Conference on Artificial Intelligence},
  year={2024}
}

Acknowledgment