SiamGAT

1. Environment setup

This code has been tested on Ubuntu 16.04, Python 3.5, Pytorch 1.2.0, CUDA 9.0. Please install related libraries before running this code:

pip install -r requirements.txt

2. Test

Dataset		SiamGAT	SiamGAT*	SiamGAT Model Link	SiamGAT* Model Link
GOT10k	AO	63.1	67.1	Google Driver/ BaiduYun(zktx)	Google Driver/ BaiduYun(d74o)
	SR0.5	74.6	78.7
	SR0.75	50.4	58.9
TrackingNet	Success	75.3	76.9	Google Driver/ BaiduYun(n2sm)	Google Driver/ BaiduYun(fxo2)
	Norm precision	80.7	82.4
	Precision	69.8	71.9
LaSOT	Success	53.9	59.5	Google Driver/ BaiduYun(dilp)
	Norm precision	63.3	69.0
	Precision	53.0	61.2
VOT2020	EAO	-	0.453	-
	A	-	0.756
	R	-	0.729
OTB100	Success	71.0	71.5	Google Driver/ BaiduYun(w1rs)	Google Driver/ BaiduYun(c6c5)
	Precision	91.7	93.0
UAV123	Success	64.6	-		-
	Precision	84.3	-

Prepare testing datasets

Download testing datasets and put them into test_dataset directory. Jsons of commonly used datasets can be downloaded from BaiduYun. If you want to test the tracker on a new dataset, please refer to pysot-toolkit to set test_dataset.

Test the tracker

python testTracker.py \    
        --config ../experiments/siamgat_googlenet_ct_alldataset/config.yaml \ # siamgat_xx_xx for SiamGAT, siamgat_ct_xx_xx for SiamGAT*
	--dataset OTB100 \                                 # dataset_name: GOT-10k, LaSOT, TrackingNet, OTB100, UAV123
	--snapshot snapshot/otb_uav_model.pth              # tracker_name

The testing result will be saved in the results/dataset_name/tracker_name directory.

3. Train

Prepare training datasets

Download the datasets：

• VID
• YOUTUBEBB (code: v7s6)
• DET
• COCO
• GOT-10K
• LaSOT
• TrackingNet

Note: training_dataset/dataset_name/readme.md has listed detailed operations about how to generate training datasets.

Download pretrained backbones

Download pretrained backbones from link and put them into pretrained_models directory.

Train a model

To train the SiamGAT model, run train.py with the desired configs:

cd tools
python train.py
	--cfg ../experiments/siamgat_googlenet/config.yaml # siamgat_xx_xx for SiamGAT, siamgat_ct_xx_xx for SiamGAT*

4. Evaluation

We provide tracking results for comparison:

• SiamGAT: BaiduYun (extract code: 8nox) or GoogleDriver.
• SiamGAT*: BaiduYun (extract code: kjym) or GoogleDriver.

If you want to evaluate the tracker on OTB100, UAV123 and LaSOT, please put those results into results directory and then run eval.py . Evaluate GOT-10k on Server. Evaluate TrackingNet on Server.

python eval.py 	                          \
	--tracker_path ./results          \ # result path
	--dataset OTB100                  \ # dataset_name
	--tracker_prefix 'otb_uav_model'   # tracker_name

5. Acknowledgement

The code is implemented based on pysot and SiamCAR. We would like to express our sincere thanks to the contributors.

6. Cite

If you use SiamGAT in your work please cite our papers:

@article{cui2022joint,
title={Joint Classification and Regression for Visual Tracking with Fully Convolutional Siamese Networks},
author={Cui, Ying and Guo, Dongyan and Shao, Yanyan and Wang, Zhenhua and Shen, Chunhua and Zhang, Liyan and Chen, Shengyong},
journal={International Journal of Computer Vision},
year={2022},
publisher={Springer},
doi = {10.1007/s11263-021-01559-4}
}

@InProceedings{Guo_2021_CVPR,
author = {Guo, Dongyan and Shao, Yanyan and Cui, Ying and Wang, Zhenhua and Zhang, Liyan and Shen, Chunhua},
title = {Graph Attention Tracking},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2021}
}

@InProceedings{Guo_2020_CVPR,
author = {Guo, Dongyan and Wang, Jun and Cui, Ying and Wang, Zhenhua and Chen, Shengyong},
title = {SiamCAR: Siamese Fully Convolutional Classification and Regression for Visual Tracking},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}