LGTrack

The implementation for the paper [Layer-Guided UAV Tracking: Enhancing Efficiency and Occlusion Robustness] on The Visual Computer

Usage

Installation

Create and activate a conda environment:

conda create -n LGTrack python=3.8
conda activate LGTrack

Install the required packages:

pip install -r requirements.txt

Data Preparation

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

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

Path Setting

Run the following command to set paths:

cd <PATH_of_LGTrack>
python tracking/create_default_local_file.py --workspace_dir . --data_dir ./data --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

python tracking/train.py --script lgtrack --config deit_tiny_patch16_224  --save_dir ./output --mode single

Testing

Put the downloaded weights on <PATH_of_LGTrack>/output/checkpoints/train/lgtrack/deit_tiny_patch16_224

python tracking/test.py lgtrack deit_tiny_patch16_224 --dataset uav123 --threads 4 --num_gpus 1
python tracking/analysis_results.py # need to modify tracker configs and names

Test FLOPs, and Params.

python tracking/profile_model.py --script lgtrack --config deit_tiny_patch16_224

Citation

If our work is useful for your research, please consider citing.