Code and results for TPAMI paper: Effective Local and Global Search for Fast Long-term Tracking

Our code has been tested on

• RTX 2080Ti GPU
• Intel i9-9900K CPU / 64 GB Memory
• Ubuntu 18.04.2 LTS
• Python3.6
• PyTorch1.2
• CUDA10.0 / cuDNN7.6

# create conda env
$ conda create --name <env_name> python=3.6
$ source activate <env_name>

# install requirements
$ conda install pytorch=1.2.0 cudatoolkit=10.0 cudnn torchvision -c pytorch
$ pip install opencv-python
$ pip install pyyaml yacs tqdm colorama matplotlib cython tensorboardX
$ pip install scikit-image

# install tensorflow and keras for reproducing the tracker '***+skim'
$ pip install tensorflow-gpu==1.14.0
$ pip install keras==2.2.5
$ conda install numpy=1.16.4 # solve "FutureWarning: Passing (type, 1) ... type is deprecated"

# install base tracker
$ cd ./modules/pysot/
$ python setup.py build_ext --inplace
$ cd ../../

# install roi_align
$ cd ./RoIAlign
$ python setup.py install
$ cd ../

# install pytorch_nms
$ cd ./pytorch_nms
$ python setup.py install
$ cd ../

• Unzip [model.zip] to ./<root_path>, [skim.zip] to ./modules/skim;

• Modify project_path and dataset_path in config_path.py.

• You can reproduce the results by running the corresponding script in ./run_<dataset> .

To fully evaluate the proposed re-detection module, we integrate our re-detection module into several trackers or replace the original re-detection module of some long-term trackers with ours. To reproduce this experiment:

• Please integrate our scripts into corresponding trackers. You can find these scripts in ./run_VOT18LT/Effectiveness_of_Re-detector/<tracker_name>.

• Download trackers: [LCT], [SiamMask], [SPLT], [DaSiam_LT & SiamVGG]

All results shown in our paper can be found in [GoogleDrive]. If you want to re-train the models, please refer to the corresponding code in ./modules

If you feel our work is useful, please cite:

@article{Zhao_TPAMI_ELGLT,
    author = {Haojie Zhao and Bin Yan and Dong Wang and Xuesheng Qian and Xiaoyun Yang and Huchuan Lu},
    title = {Effective Local and Global Search for Fast Long-term Tracking},
    journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
    year = {2022}
}

If you have any questions, you can contact me by email.