[TGRS2023] SmallTrack: Wavelet Pooling and Graph Enhanced Classification for UAV Small Object Tracking

This is an official pytorch implementation of the 2023 IEEE Transactions on Geoscience and Remote Sensing paper:

SmallTrack: Wavelet Pooling and Graph Enhanced Classification for UAV Small Object Tracking
(accepted by IEEE Transactions on Geoscience and Remote Sensing, DOI: 10.1109/TGRS.2023.3305728)

The paper can be downloaded from IEEE Xplore

The models and raw results can be downloaded from [GitHub] or [BaiduYun].

The tracking demos are displayed on the Bilibili

Proposed modules

• DWT in backbone

• GEM in model_builder

UAV Tracking

Datasets	smalltrack_r50_l234
UAV20L(Suc./Pre.)	0.600/0.797
UAVDT(Suc./Pre.)	0.637/0.866
DTB70(Suc./Pre.)	0.654/0.858
VisDrone2019-SOT-test-dev(Suc./Pre.)	0.625/0.849
LaTOT(Suc./Pre./N.Pre.)	0.271/0.438/0.339

Note:

• r50_lxyz denotes the outputs of stage x, y, and z in ResNet-50.
• The suffixes DTB70 is designed for the DTB70, the default (without suffix) is designed for UAV20L and UAVDT.
• e20 in parentheses means checkpoint_e20.pth

Installation

Please find installation instructions in INSTALL.md.

Quick Start: Using SmallTrack

Add SmallTrack to your PYTHONPATH

export PYTHONPATH=/path/to/smalltrack:$PYTHONPATH

demo

python tools/demo.py \
    --config experiments/smalltrack_r50_l234/config.yaml \
    --snapshot experiments/smalltrack_r50_l234/checkpoint_e20.pth
    --video demo/bag.avi

Download testing datasets

Download datasets and put them into testing_dataset directory. Jsons of commonly used datasets can be downloaded from Google Drive or BaiduYun. If you want to test tracker on new dataset, please refer to pysot-toolkit to setting testing_dataset.

Test tracker

• Note that it is not necessary to generate the json files for the test dataset as per pysot.
• We read the dataset format online to generate the corresponding dictionary, the relevant files are in visdrone.py

cd experiments/smalltrack_r50_l234
python -u ../../tools/test.py 	\
	--snapshot checkpoint_e20.pth 	\ # model path
	--dataset UAV20L 	\ # dataset name
	--config config.yaml	  # config file

The testing results will in the current directory(results/dataset/model_name/)

Eval tracker

assume still in experiments/smalltrack_r50_l234

python ../../tools/eval.py 	 \
	--tracker_path ./results \ # result path
	--dataset UAV20L        \ # dataset name
	--num 1 		 \ # number thread to eval
	--tracker_prefix 'ch*'   # tracker_name

Training 🔧

• training datasets and corresponding Json files are provided in [Baidu Drive] for COCO, DET, GOT-10k, LaSOT, VID, YouTube-BoundingBoxes
• See TRAIN.md for detailed instruction.

Acknowledgement

The code based on the PySOT , SiamBAN , GAL , WaveCNets and Wavelet-Attention We would like to express our sincere thanks to the contributors.

Citation:

If you find this work useful for your research, please cite the following papers:

@ARTICLE{10220112,
  author={Xue, Yuanliang and Jin, Guodong and Shen, Tao and Tan, Lining and Wang, Nian and Gao, Jing and Wang, Lianfeng},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={SmallTrack: Wavelet Pooling and Graph Enhanced Classification for UAV Small Object Tracking}, 
  year={2023},
  volume={61},
  pages={1-15},
  keywords={Target tracking;Object tracking;Wavelet transforms;Feature extraction;Task analysis;Remote sensing;Visualization;Aerial tracking;graph enhanced classification;remote sensing;Siamese neural network;wavelet pooling layer (WPL)},
  doi={10.1109/TGRS.2023.3305728}}

If you have any questions about this work, please contact with me via [email protected]