Home

Awesome

CTTrack

The official implementation of the AAAI 2023 Oral paper <u> [Compact Transformer Tracker with Correlative Masked Modeling] </u>

[Models and Raw results] (Google Driver) [Models and Raw results] (Baidu Driver: LRBW)

backbone

News

[ Jan 18, 2023]

[ Jan 11, 2023]

[ Nov 19, 2022]

Strong performance

TrackerUAV123 (P)LaSOT (NP)TrackingNet (AUC)GOT-10k (AO)
CTTrack93.379.784.972.8
MixFormer* (CVPR2022)91.079.983.970.7
CSWinTT* (CVPR2022)90.375.281.969.4
UTT* (CVPR2022)--79.767.2
STARK* (ICVV2021)-77.082.068.8
TransT* (CVPR2021)87.673.881.467.1
TrDiMP* (CVPR2021)87.673.278.468.8
STMTrack* (CVPR2021)-69.380.364.2
AutoMatch* (ICVV2021)83.867.576.065.2
SiamGAT* (CVPR2021)84.363.3-62.7
KYS* (ECCV2020)-63.374.063.6
SiamAttn* (CVPR2020)84.564.875.2-
SiamFC++* (AAAI2020)80.462.375.459.5
SiamRPN++* (CVPR2019)84.056.973.351.7
DiMP* (ICCV2019)84.966.474.061.1
ATOM* (CVPR2019)82.757.670.355.6

Install the environment

conda create -n cttrack python=3.7
conda activate cttrack
pip install -r requirements.txt

Data Preparation

Put the tracking datasets in ./data. It should look like:

${CTTRACK_ROOT}
 -- data
     -- lasot
         |-- airplane
         |-- basketball
         |-- bear
         ...
     -- got10k
         |-- test
         |-- train
         |-- val
     -- trackingnet
         |-- TRAIN_0
         |-- TRAIN_1
         ...
         |-- TRAIN_11
         |-- TEST

Set project paths

Run the following command to set paths for this project

python tracking/create_default_local_file.py --workspace_dir . --data_dir ./data --save_dir .

After running this command, you can also modify paths by editing these two files

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

Train CTTrack

Train CTTrack-B

python tracking/train.py --script cttrack --config baseline --save_dir . --mode single
python tracking/train.py --script cttrack_online --config baseline --save_dir . --mode single --script_prv cttrak --config_prv baseline  

Train CTTrack-L

python tracking/train.py --script cttrack --config baseline_L --save_dir . --mode single
python tracking/train.py --script cttrack_online --config baseline_L --save_dir . --mode single --script_prv cttrak --config_prv baseline  

Test and evaluate CTTrack on benchmarks

Test CTTrack-B

python tracking/test.py cttrack baseline --dataset otb --threads 32
python tracking/test.py cttrack baseline --dataset uav --threads 32
python tracking/test.py cttrack baseline --dataset lasot --threads 32
python tracking/test.py cttrack baseline --dataset got10k_test --threads 32
python tracking/test.py cttrack baseline --dataset trackingnet --threads 32

Test CTTrack-L

python tracking/test.py cttrack baseline_L --dataset otb --threads 32
python tracking/test.py cttrack baseline_L --dataset uav --threads 32
python tracking/test.py cttrack baseline_L --dataset lasot --threads 32
python tracking/test.py cttrack baseline_L --dataset got10k_test --threads 32
python tracking/test.py cttrack baseline_L --dataset trackingnet --threads 32

Evaluate CTTrack

LaSOT/GOT10k-test/TrackingNet/OTB100/UAV123

python tracking/analysis_results.py {script}  {config}  {dataset_name}

For example

python tracking/analysis_results.py cttrack baseline trackingnet

VOT2020

Before evaluating "CTTrack" on VOT2020, please install VOT toolkit which is required to evaluate our tracker. To download and install VOT toolkit, you can follow this tutorial. For convenience, you can use our example workspaces of VOT toolkit under external/vot20/ by setting trackers.ini.

cd external/vot20/<workspace_dir>
vot evaluate --workspace . CTTrackPython
# generating analysis results
vot analysis --workspace . CTTrackPython --nocache

Visualize attention maps

attention

Model Zoo and raw results

[Models and Raw results] (Google Driver) [Models and Raw results] (Baidu Driver: RLBW)

Contact

Zikai Song: skyesong@hust.edu.cn

Run Luo: lr_8823@hust.edu.cn

Acknowledgments

Citation

If you think this project is helpful, please feel free to leave a star⭐️ and cite our paper:

@InProceedings{Song_2023_AAAI,
   author    = {Song, Zikai and Luo, Run and Yu, Junqing and Chen, Yi-Ping Phoebe and Yang, Wei},
   title     = {Compact Transformer Tracker with Correlative Masked Modeling},
   booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)},
   month     = {February},
   year      = {2023}
}