ReFocusTIRTracking

The official implementation for the [TNNLS 2024] paper: "Refocus the Attention for Parameter-Efficient Thermal Infrared Object Tracking".

:rocket: Update Models and Results (2024/07/17)
Models & Raw Results & Training logs [Google Driver]
Models & Raw Results & Training logs [Baidu Driver: v27s]

Install the environment

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 ./output

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

Data Preparation

prepare the LSOTB dataset. It should look like:

${PROJECT_ROOT}
 -- LSOTB
     -- Train
         |-- TrainingData
         |-- MyAnnotations
         ...
     -- Eval
         |-- aircraft_car
         |-- airplane_H_001
         |-- LSOTB-TIR-120.json
         |-- LSOTB-TIR-136.json
         |-- LSOTB-TIR-LT11.json
         |-- LSOTB-TIR-ST100.json

Training

Download pre-trained OSTrack_ep0300.pth.tar from above driver link and put it under $PROJECT_ROOT$/pretrained_models . Then

bash xtrain.sh

Replace --config with the desired model config under experiments/refocus. We use wandb to record detailed training logs, in case you don't want to use wandb, set --use_wandb 0.

Evaluation

Download the model weights OSTrack_ep0060.pth.tar from above driver link

Download the evaluation toolkits from here

Put the downloaded weights on $PROJECT_ROOT$/checkpoints/

Change the corresponding values of lib/test/evaluation/local.py to the actual benchmark saving paths. Then

bash ytest.sh

Acknowledgments

• This repo is based on OSTrack which is an excellent work.

Contact

If you have any question, feel free to email laisimiao@mail.dlut.edu.cn. ^_^