chaoma99 / HCFTstar

Robust Visual Tracking via Hierarchical Convolutional Features

Home Page:https://sites.google.com/site/chaoma99/hcft-tracking

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Robust Visual Tracking Via Hierarchical Convolutional Features

Introduction

This is the research code of the TPAMI 2019 paper:

Chao Ma, Jia-Bin Huang, Xiaokang Yang and Ming-Hsuan Yang, "Robust Visual Tracking via Hierarchical Convolutional Features", IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2019.

This work is the extended version of the ICCV 2015 paper:

Chao Ma, Jia-Bin Huang, Xiaokang Yang and Ming-Hsuan Yang, "Hierarchical Convolutional Features for Visual Tracking", ICCV 2015. For the early conference version code, please visit our Project page.

In this source code, we added the components for (i) scale estimation and (ii) target redetection from tracking failures caused by heavy occlusion or targets moving out of the view. We propose to use another correlation filter to maintain a long-term memory of target appearance as a classifier. By adjusting the parameters, we tailor the off-the-shelf EdgeBox toolbox to generate two types of region proposals: (i) proposals tightly around the estimated location as candidates for scale estimation; (ii) proposals sampled across the whole image as candidates for target re-detection. We apply the classifier to these two types of proposals and respectively select the proposals with highest response scores for scale estimation and target re-detection.

The correlation filters with convolutional features is a state-of-the-art tracker that exploits rich feature hierarchy from deep convolutional neural networks for visual tracking. For more details, please visit our Project page.

Start

  1. Setup matconvet (recompile it if it is not comptiable with your machine)

  2. cd vgg_model and run download.sh to download the VGG-19 model "imagenet-vgg-verydeep-19.mat"

  3. run "run_tracker_demo.m"

Note that the code for visualising CNN layers is in the folder "visualization"

Citation

If you find the code and dataset useful in your research, please consider citing:

@article{Ma-HCFTstar-2017,
  title={Robust visual tracking via hierarchical convolutional features},
  author={Ma, Chao and Huang, Jia-Bin and Yang, Xiaokang and Yang, Ming-Hsuan},
  journal={IEEE transactions on pattern analysis and machine intelligence},
  volume={41},
  number={11},
  pages={2709--2723},
  year={2019}
}

@inproceedings{Ma-ICCV-2015,
    title={Hierarchical Convolutional Features for Visual Tracking},
    Author = {Ma, Chao and Huang, Jia-Bin and Yang, Xiaokang and Yang, Ming-Hsuan},
    booktitle = {Proceedings of the IEEE International Conference on Computer Vision},
    pages={},
    Year = {2015}
}

Contents

Folder description

Feedbacks and comments are welcome! Feel free to contact us via chaoma99@gmail.com or jbhuang@vt.edu.

Enjoy!

About

Robust Visual Tracking via Hierarchical Convolutional Features

https://sites.google.com/site/chaoma99/hcft-tracking


Languages

Language:MATLAB 41.8%Language:HTML 19.9%Language:Cuda 16.1%Language:C++ 12.3%Language:TeX 4.2%Language:Python 2.7%Language:C 1.5%Language:Shell 0.9%Language:Makefile 0.5%Language:CSS 0.2%Language:M 0.0%Language:JavaScript 0.0%Language:Mathematica 0.0%