The official implementation for the CVPR 2026 paper
[Toward Low-Cost yet Effective Temporal Learning for UAV Tracking]
Models, Raw Results, and Training Logs are available for download at here, code: 6n35
Put the training 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
For ease of testing, we have made the structured dataset available for download at here, code: e22r
Put the test datasets in ./data. It should look like:
${PROJECT_ROOT}
-- data
-- UAV123
|-- anno
|-- data_seq
-- UAV123_10fps
|-- anno
|-- data_seq
-- uavdt
|-- anno
|-- sequences
-- V4RFlight112
|-- anno
|-- anno_l
|-- data_seq
|-- attributes
-- DTB70
|-- Animal1
|-- Animal2
...
-- VisDrone2018-SOT-test-dev
|-- annotations
|-- sequences
|-- attributes
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
Download pre-trained DeiT-tiny distilled weights and rename as deit_distilled.pth and put it under $PROJECT_ROOT$/pretrained_models
python tracking/train.py \
--script letrack --config baseline_WOCE \
--save_dir ./output \
--mode multiple --nproc_per_node 4 \
--use_wandb 0
Replace --config with the desired model config under experiments/letrack.
We use wandb to record detailed training logs, in case you don't want to use wandb, set --use_wandb 0.
- UAV123 or other off-line evaluated benchmarks (modify
--datasetcorrespondingly)
python tracking/test.py --tracker_param letrack --dataset uav123 --threads 8 --num_gpus 4
python tracking/analysis_results.py # need to modify tracker configs and names
- uav123_10fps
python tracking/test.py --tracker_param letrack --dataset uav123_10fps --threads 8 --num_gpus 4
- uavtrack_L
python tracking/test.py --tracker_param letrack --dataset uavtrack --threads 8 --num_gpus 4
- uavtrack112
python tracking/test.py --tracker_param letrack --dataset uavtrack112 --threads 8 --num_gpus 4
- uavdt
python tracking/test.py --tracker_param letrack --dataset uavdt --threads 8 --num_gpus 4
- dtb70
python tracking/test.py --tracker_param letrack --dataset dtb70 --threads 8 --num_gpus 4
- visdrone
python tracking/test.py --tracker_param letrack --dataset visdrone --threads 8 --num_gpus 4
Note: The speeds reported in our paper were tested on a single RTX2080Ti GPU.
python tracking/profile_model.py
For any questions or cooperation, please contact xcc23cg@163.com or wechat: chaocan23
If our work is useful for your research, please consider citing:
@inproceedings{letrack,
title={Toward Low-Cost yet Effective Temporal Learning for UAV Tracking},
author={Xue, Chaocan and Liang, Qihua and Zhong, Bineng and Zu, Yanting and Xue, Yuanliang and Xia, Haiying and Song, Shuxiang},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={42538--42548},
year={2026}
}Friendly link: SGLATrack (CVPR 2025)