lql-team / TSDM

a RGB-D tracker based on CNN with using depth information

TSDM

A RGB-D tracker base on CNN with using depth information.
Download paper https://arxiv.org/abs/2005.04063
The code is simple here.

Requirment Libs

• Conda with Python 3.7.
• Nvidia GPU.
• PyTorch 1.0
• OpenCV

Requirment models

• Res20.pth: retrained SiamRPN++ model (works with M-g)
• modelRes: not retrained SiamRPN++ model
• High-Low-two.pth: D-r model
  Pease put the three models into data_and_result/weight. Link as follow:
  BAIDU YUN: https://pan.baidu.com/s/1Z2c9SymPIRTA_-4p5W1hHA pin: faon
  Google Drive: https://drive.google.com/drive/folders/17EN9IU-GOhFQt7middHVaNQFwWj7U8MP?usp=sharing

Requirment datasets

• VOT-RGBD2019: www.votchallenge.net/
• PTB dataset: http://tracking.cs.princeton.edu/dataset.html
• demo dataset: https://drive.google.com/drive/folders/19O6o8H_CblZehAJpNQ_KIFe2XGVMs_eY?usp=sharing
  Please put three datasets above into "data_and_result/test_data/". If you only run demo, you can only download demo dataset and put it into "data_and_result/test_data/"

Test demo

If you wang to run the demo, please run "python3 test.py --sequence n". Here, n ranges {1,2,3}.

Acknowlege

Thanks to SiamRPN++, it is the core of our model. Paper in https://arxiv.org/pdf/1812.11703.pdf

Others

I will always update and maintain this repository.