Aba-ViTrack

The official implementation of the ICCV 2023 paper Adaptive and Background-Aware Vision Transformer for Real-Time UAV Tracking

Install the environment

This code has been tested on Ubuntu 18.04, CUDA 10.2. Please install related libraries before running this code:

conda create -n abavitrack python=3.8
conda activate abavitrack
bash install.sh

Model and raw results

The trained model and the raw tracking results are provided in the Baidu Netdisk(code: nen9) or Google Drive.

Run demo

Download the model and put it in checkpoints

python demo.py --initial_bbox 499 421 102 179

Run Aba-ViTrack on custom video

Update tracker using ground truth

python run_eval.py --data_path /path/to/video/file --save_path /path/to/saving/dir --update_rate 10

Update tracker using YOLO as detector

python track_by_detect.py --data_path /path/to/video/file --save_path /path/to/saving/dir --update_rate 10 --det_weights /path/to/yolo/weights

📝 By running these codes, you can run your tracker on a custom video and store the results in xyxy format. These codes are used to evaluate your tracker using detection methods to find its effect on detection performance.

Citation

@InProceedings{Li_2023_ICCV,
    author    = {Li, Shuiwang and Yang, Yangxiang and Zeng, Dan and Wang, Xucheng},
    title     = {Adaptive and Background-Aware Vision Transformer for Real-Time UAV Tracking},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {13989-14000}
}

setarekhosravi/Aba-ViTrack