# AlphaRefine **Repository Path**: jankin987/AlphaRefine ## Basic Information - **Project Name**: AlphaRefine - **Description**: No description available - **Primary Language**: Unknown - **License**: GPL-3.0 - **Default Branch**: master - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 0 - **Created**: 2021-04-02 - **Last Updated**: 2023-04-10 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README # Alpha-Refine This is the official implementation of [Alpha-Refine: Boosting Tracking Performance by Precise Bounding Box Estimation ](https://arxiv.org/abs/2012.06815). ![Architecture](doc/asset/AR-Architecture.png) ## News - :warning: We provide a concise script [demo.py](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](https://github.com/ymzis69/AlphaRefine_TensorRT). - 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** - :trophy: **Alpha-Refine wins VOT2020 Real-Time Challenge with EAOMultistart 0.499!** - VOT2020 winner presentation [slide](VOT20-RT-Report.pdf) has been uploaded. ## Setup Alpha-Refine * **Install AlphaRefine** ```bash 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`. | Tracker | Backbone | Latency | AUC(%) | Model | |:--------------:|:----------------:|:-----------:|:-----------:|:----------------:| | AR34_c+m | ResNet34 | 5.1ms | 55.9 | [google](https://drive.google.com/file/d/1drLqNq4r9g4ZqGtOGuuLCmHJDh20Fu1m/view?usp=sharing)/[baidu](https://pan.baidu.com/s/1ZCJKk1mXE_96BEpwGiEuMQ)[key:jl1m]| | AR18_c+m | ResNet18 | 4.2ms | 55.0 | [google](https://drive.google.com/file/d/1ANf0KCvlFBbGQPpvT-3WNiy414ANkgLZ/view?usp=sharing)/[baidu](https://pan.baidu.com/s/1IIaRNkFVPSG1s71g255CHw)[key:83ef]| 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](#performance) * **Demo** We provide a concise [demo.py](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](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](demo.py) as an example for applying alpha refine to dimp. ## How to Train Alpha-Refine Please refer to [doc/TRAIN.md](doc/TRAIN.md) for the guidance of training Alpha-Refine. After training, you can refer to [doc/Reproduce.md](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](https://arxiv.org/abs/2012.06815). **You can refer to [doc/Reproduce.md](doc/Reproduce.md) for reproducing our result.** * **LaSOT** | Tracker | Success Score | Speed (fps) | Paper/Code | |:----------- |:----------------:|:----------------:|:----------------:| | ARDiMP (ours) | 0.654 | 32 (RTX 2080Ti) | [Paper](https://arxiv.org/abs/2012.06815)/[Result](https://drive.google.com/file/d/1UNPwz7qP8SeBTxHF_Cw0JLmrN1jTqJJE/view?usp=sharing) | | Siam R-CNN (CVPR20) | 0.648 | 5 (Tesla V100) | [Paper](https://arxiv.org/pdf/1911.12836.pdf)/[Code](https://github.com/VisualComputingInstitute/SiamR-CNN) | | DimpSuper | 0.631 | 39 (RTX 2080Ti) | [Paper](https://arxiv.org/pdf/2003.12565.pdf)/[Code](https://github.com/visionml/pytracking) | | ARDiMP50 (ours) | 0.602 | 46 (RTX 2080Ti) | [Paper](https://arxiv.org/abs/2012.06815)/[Result](https://drive.google.com/file/d/1wJc_-1lCxeGlqEAKd1qER1x_4bWAhujv/view?usp=sharing) | | PrDiMP50 (CVPR20) | 0.598 | 30 (Unkown GPU) | [Paper](https://arxiv.org/pdf/2003.12565.pdf)/[Code](https://github.com/visionml/pytracking) | | LTMU (CVPR20) | 0.572 | 13 (RTX 2080Ti) | [Paper](https://arxiv.org/abs/2004.00305)/[Code](https://github.com/Daikenan/LTMU) | | DiMP50 (ICCV19) | 0.568 | 59 (RTX 2080Ti) | [Paper](https://arxiv.org/pdf/1904.07220.pdf)/[Code](https://github.com/visionml/pytracking) | | Ocean (ECCV20) | 0.560 | 25 (Tesla V100) | [Paper](https://arxiv.org/abs/2006.10721)/[Code](https://github.com/researchmm/TracKit) | | ARSiamRPN (ours) | 0.560 | 50 (RTX 2080Ti) | [Paper](https://arxiv.org/abs/2012.06815)/[Result](https://drive.google.com/file/d/1u-ou43O_RU9oRFx1UKjzeYe6e-4qnMZZ/view?usp=sharing) | | SiamAttn (CVPR20) | 0.560 | 45 (RTX 2080Ti) | [Paper](https://arxiv.org/pdf/2004.06711.pdf)/[Code]() | | SiamFC++GoogLeNet (AAAI20)| 0.544 | 90 (RTX 2080Ti) | [Paper](https://arxiv.org/pdf/1911.06188.pdf)/[Code](https://github.com/MegviiDetection/video_analyst) | | MAML-FCOS (CVPR20) | 0.523 | 42 (NVIDIA P100) | [Paper](https://arxiv.org/pdf/2004.00830.pdf)/[Code]() | | GlobalTrack (AAAI20) | 0.521 | 6 (GTX TitanX) | [Paper](https://arxiv.org/abs/1912.08531)/[Code](https://github.com/huanglianghua/GlobalTrack) | | ATOM (CVPR19) | 0.515 | 30 (GTX 1080) | [Paper](https://arxiv.org/pdf/1811.07628.pdf)/[Code](https://github.com/visionml/pytracking) | ## Acknowledgments * This repo is based on [Pytracking](https://github.com/visionml/pytracking.git) which is an exellent work. * Thanks for [pysot](https://github.com/STVIR/pysot) and [RTMDNet](https://github.com/IlchaeJung/RT-MDNet) from which we borrow the code as base trackers.