The official implementation of the ICCV 2023 paper Adaptive and Background-Aware Vision Transformer for Real-Time UAV Tracking
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
The trained model and the raw tracking results are provided in the Baidu Netdisk(code: nen9) or Google Drive.
Download the model and put it in checkpoints
python demo.py --initial_bbox 499 421 102 179
python run_eval.py --data_path /path/to/video/file --save_path /path/to/saving/dir --update_rate 10
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.
@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}
}