This project is mainly based on HelloRicky123/Siamese-RPN. What I've done is just add a Feature Pyramid Network method to the original AlexNet structures.

For more details about siameseRPN please refer to the paper : High Performance Visual Tracking with Siamese Region Proposal Network by Bo Li, Junjie Yan,Wei Wu, Zheng Zhu, Xiaolin Hu.

For more details about Feature Pyramid Network please refer to the paper: Feature Pyramid Network for Object Detection by Tsung-Yi Lin, Piotr Dollár, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie.

• Siamese Region Proposal Networks

  

• Feature Pyramid Networks

  

• SimaeseRPN+FPN

  • Template Branch

    

  • Detection Branch

    

This project can get 0.618 AUC on OTB100, which also achieves overall 1.3% progress than the performance of baseline Siamese-RPN. Additionally, based on the ablation study results, it also shows that it can achieve robust performance different operating systems and GPUs.

I only use pre-trained models to finish my experiments，so here I would post the testing dataset OTB100 I get from http://cvlab.hanyang.ac.kr/tracker_benchmark/

If you don't want to download through the website above, you can just download: https://pan.baidu.com/s/1vWIn8ovCGKmlgIdHdt_MkA key: p8u4

For more details about OTB100 please refer to the paper: Object Tracking Benchmark by Yi Wu, Jongwoo Lim, Ming-Hsuan Yang.

I didn't do any training but I still keep the baseline training method in my project. So if you have VID dataset or youtube-bb dataset, I would just post the steps of training here

Create dataset:

python bin/create_dataset_ytbid.py --vid-dir /PATH/TO/ILSVRC2015 --ytb-dir /PATH/TO/YT-BB --output-dir /PATH/TO/SAVE_DATA --num_threads 6

Create lmdb:

python bin/create_lmdb.py --data-dir /PATH/TO/SAVE_DATA --output-dir /PATH/TO/RESULT.lmdb --num_threads 12

Train:

python bin/train_siamrpn.py --data_dir /PATH/TO/SAVE_DATA

If want to test the tracker, please first change the project path:

sys.path.append('[your_project_path]')

And then choose the combinations of different layers I putted in the net/network.py

then input your model path and dataset path to run:

python bin/test_OTB.py -ms [your_model_path] -v tb100 -d [your_dataset_path]

I've exported my anaconda and pip environment into /env/conda_env.yaml and /env/pip_requirements.txt

if you want to use it, just run the command below accordingly

for anaconda:

conda create -n [your_env_name] -f conda_env.yaml

for pip:

pip install -r requirements.txt

Model which the baseline uses: https://pan.baidu.com/s/1vSvTqxaFwgmZdS00U3YIzQ keyword: v91k

Model after training 50 epoch: https://pan.baidu.com/s/1m9ISra0B04jcmjW1n73fxg keyword: 0s03

(1)

DELL-Precision-7530

OS: Ubuntu 18.04 LTS CPU: Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz

Memory: 2*8G DDR4 2666MHZ

GPU: Nvidia Quadro P1000

(2)

HP OMEN

OS: Windows 10 Home Edition

CPU: Intel(R) Core(TM) i7-9750H CPU @ 2.6GHz

Memory: 2*8G DDR4 2666MHZ

GPU: Nvidia Geforce RTX2060

On Ubuntu and Quadro P1000

• AUCs with model siamrpn_38.pth

• AUCs with model siamrpn_50.pth

On Windows 10 and Nvidia Geforce RTX2060

• AUCs with model siamrpn_38.pth

• AUCs with model siamrpn_50.pth

[1] B. Li, J. Yan, W. Wu, Z. Zhu, X. Hu, High Performance Visual Tracking with Siamese Region Proposal Network, inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pages 8971-8980.

[2] T. Lin, P. Dollar, R. Girshick, K. He, B. Hariharan, S. Belongie, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pages 2117-2125.

[3] Y. Wu, J. Lim, M. Yang, "Object Tracking Benchmark", in IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, pages 1834-1848.