🏆News: We won the VOT-18 real-time challenge

🏆News: We won the second place in the VOT-18 long-term challenge

DaSiamRPN

This repository includes PyTorch code for reproducing the results on VOT2018.

Distractor-aware Siamese Networks for Visual Object Tracking

Zheng Zhu^*, Qiang Wang^*, Bo Li^*, Wei Wu, Junjie Yan, and Weiming Hu

European Conference on Computer Vision (ECCV), 2018

Introduction

SiamRPN formulates the task of visual tracking as a task of localization and identification simultaneously, initially described in an CVPR2018 spotlight paper. (Slides at CVPR 2018 Spotlight)

DaSiamRPN improves the performances of SiamRPN by (1) introducing an effective sampling strategy to control the imbalanced sample distribution, (2) designing a novel distractor-aware module to perform incremental learning, (3) making a long-term tracking extension. ECCV2018. (Slides at VOT-18 Real-time challenge winners talk)

Prerequisites

CPU: Intel(R) Core(TM) i7-7700 CPU @ 3.60GHz GPU: NVIDIA GTX1060

• python2.7
• pytorch == 0.3.1
• numpy
• opencv

Pretrained model for SiamRPN

In our tracker, we use an AlexNet variant as our backbone, which is end-to-end trained for visual tracking. The pretrained model can be downloaded from google drive: SiamRPNBIG.model. Then, you should copy the pretrained model file SiamRPNBIG.model to the subfolder './code', so that the tracker can find and load the pretrained_model.

Detailed steps to install the prerequisites

• install pytorch, numpy, opencv following the instructions in the run_install.sh. Please do not use conda to install.
• you can alternatively modify /PATH/TO/CODE/FOLDER/ in tracker_SiamRPN.m If the tracker is ready, you will see the tracking results. (EAO: 0.3827)

Results

All results can be downloaded from Google Drive.

Demo and Test on OTB2015

• To reproduce the reuslts on paper, the pretrained model can be downloaded from Google Drive: SiamRPNOTB.model.
  ⚡ ⚡ This model is the fastest (~200fps) Siamese Tracker with AUC of 0.655 on OTB2015. ⚡ ⚡

• You must download OTB2015 dataset (download script) at first.

A simple test example.

cd code
python demo.py

If you want to test the performance on OTB2015, please using the follwing command.

cd code
python test_otb.py
python eval_otb.py OTB2015 "Siam*" 0 1

License

Licensed under an MIT license.

Citing DaSiamRPN

If you find DaSiamRPN and SiamRPN useful in your research, please consider citing:

@inproceedings{Zhu_2018_ECCV,
  title={Distractor-aware Siamese Networks for Visual Object Tracking},
  author={Zhu, Zheng and Wang, Qiang and Bo, Li and Wu, Wei and Yan, Junjie and Hu, Weiming},
  booktitle={European Conference on Computer Vision},
  year={2018}
}

@InProceedings{Li_2018_CVPR,
  title = {High Performance Visual Tracking With Siamese Region Proposal Network},
  author = {Li, Bo and Yan, Junjie and Wu, Wei and Zhu, Zheng and Hu, Xiaolin},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2018}
}