dman_mot's Introduction

Online Multi-Object Tracking with DMANs

This is the implementation of our ECCV 2018 paper Online Multi-Object Tracking with Dual Matching Attention Networks. We integrate the ECO [1] for single object tracking. The code framework for MOT benefits from the MDP [2].

Prerequisites

Cuda 8.0
Cudnn 5.1
Python 2.7
Keras 2.0.5
Tensorflow 1.1.0

For example:

conda create -n mot anaconda python=2.7
conda activate mot
conda install -c menpo opencv
pip install tensorflow-gpu==1.1.0
pip install keras==2.0.5

Usage

Download the DMAN model and put it into the "model/" folder.
Download the MOT16 dataset, unzip it to the "data/" folder.
Cd to the "ECO/" folder, run the script install.m to compile libs for the ECO tracker
Run the socket server script:

python calculate_similarity.py

Run the socket client script DMAN_demo.m in Matlab.

Citation

If you use this code, please consider citing:

@inproceedings{zhu-eccv18-DMAN,
    author    = {Zhu, Ji and Yang, Hua and Liu, Nian and Kim, Minyoung and Zhang, Wenjun and Yang, Ming-Hsuan},
    title     = {Online Multi-Object Tracking with Dual Matching Attention Networks},
    booktitle = {European Computer Vision Conference},
    year      = {2018},
}

References

[1] Danelljan, M., Bhat, G., Khan, F.S., Felsberg, M.: ECO: Efficient convolution operators for tracking. In: CVPR (2017)

[2] Xiang, Y., Alahi, A., Savarese, S.: Learning to track: Online multi-object tracking by decision making. In: ICCV (2015)

dman_mot's People

Contributors

Stargazers

Watchers

dman_mot's Issues

There is no DMAN model can download

https://github.com/jizhu1023/DMAN_MOT/blob/master/http
Page not found

training code?

Will you push the training code to this repo, or when? Thank you~

Errors in DMAN_demo.m

Hello, thanks for your works. When I run the DMAN_demo.m, there are some errors:
1、In MOT_track.m, the function MDP_value is lost

2、I copy the MDP_value.m from MDP project, however, there are too many parameters input to the MDP_value function.