Alpha-Refine

This is the official implementation of Alpha-Refine: Boosting Tracking Performance by Precise Bounding Box Estimation .

News

• ⚠️ We provide a concise script demo.py as an example of applying alpha refine to dimp. We recommend taking this script as the starting point of exploring our project.

• A TensorRT optimized version of AlphaRefine is available here.

• The code for CVPR2021 is updated. The old version is still available by

    git clone -b vot2020 https://github.com/MasterBin-IIAU/AlphaRefine.git 
  

• AlphaRefine is accepted by the CVPR2021

• 🏆 Alpha-Refine wins VOT2020 Real-Time Challenge with EAOMultistart 0.499!

• VOT2020 winner presentation slide has been uploaded.

Setup Alpha-Refine

• Install AlphaRefine

git clone https://github.com/MasterBin-IIAU/AlphaRefine.git
cd AlphaRefine

Run the installation script to install all the dependencies. You need to provide the ${conda_install_path} (e.g. ~/anaconda3) and the name ${env_name} for the created conda environment (e.g. alpha).

# install dependencies
bash install.sh ${conda_install_path} ${env_name}
conda activate alpha
python setup.py develop

• Download AlphaRefine Models

We provide the models of AlphaRefine here. The AUC and Latency are tested with SiamRPN++ as the base tracker on LaSOT dataset, using a RTX 2080Ti GPU.

We recommend download the model into ltr/checkpoints/ltr/SEx_beta.

When combined with more powerful base trackers, AlphaRefine leads to very competitive tracking systems (e.g. ARDiMP). Following are some of the best performed trackers on LaSOT. Results are present in Performance

• Demo

We provide a concise demo.py as an example for applying alpha refine to dimp. We recommend you should take this script as the starting point of exploring our project. You may need doc/Reproduce.md for setting up the base trackers of our experiments.

How to apply Alpha-Refine to Your Own Tracker

We provide a concise demo.py as an example for applying alpha refine to dimp.

How to Train Alpha-Refine

Please refer to doc/TRAIN.md for the guidance of training Alpha-Refine.

After training, you can refer to doc/Reproduce.md for reproducing our experiment result.

Performance

When combined with more powerful base trackers, AlphaRefine leads to very competitive tracking systems (e.g. ARDiMP). For more performance reports, please refer to our paper. You can refer to doc/Reproduce.md for reproducing our result.

• LaSOT

  Tracker	Success Score	Speed (fps)	Paper/Code
  ARDiMP (ours)	0.654	32 (RTX 2080Ti)	Paper/Result
  Siam R-CNN (CVPR20)	0.648	5 (Tesla V100)	Paper/Code
  DimpSuper	0.631	39 (RTX 2080Ti)	Paper/Code
  ARDiMP50 (ours)	0.602	46 (RTX 2080Ti)	Paper/Result
  PrDiMP50 (CVPR20)	0.598	30 (Unkown GPU)	Paper/Code
  LTMU (CVPR20)	0.572	13 (RTX 2080Ti)	Paper/Code
  DiMP50 (ICCV19)	0.568	59 (RTX 2080Ti)	Paper/Code
  Ocean (ECCV20)	0.560	25 (Tesla V100)	Paper/Code
  ARSiamRPN (ours)	0.560	50 (RTX 2080Ti)	Paper/Result
  SiamAttn (CVPR20)	0.560	45 (RTX 2080Ti)	Paper/Code
  SiamFC++GoogLeNet (AAAI20)	0.544	90 (RTX 2080Ti)	Paper/Code
  MAML-FCOS (CVPR20)	0.523	42 (NVIDIA P100)	Paper/Code
  GlobalTrack (AAAI20)	0.521	6 (GTX TitanX)	Paper/Code
  ATOM (CVPR19)	0.515	30 (GTX 1080)	Paper/Code

Acknowledgments

• This repo is based on Pytracking which is an exellent work.
• Thanks for pysot and RTMDNet from which we borrow the code as base trackers.