Learning to Track: Online Multi-Object Tracking by Decision Making
Created by Yu Xiang at CVGL, Stanford University.
Introduction
MDP_Tracking is a online multi-object tracking framework based on Markov Decision Processes (MDPs).
http://cvgl.stanford.edu/projects/MDP_tracking/
License
MDP_Tracking is released under the MIT License (refer to the LICENSE file for details).
Citation
If you find MDP_Tracking useful in your research, please consider citing:
@inproceedings{xiang2015learning,
Author = {Xiang, Yu and Alahi, Alexandre and Savarese, Silvio},
Title = {Learning to Track: Online Multi-Object Tracking by Decision Making},
Booktitle = {International Conference on Computer Vision (ICCV)},
Year = {2015}
}
Usage on the 2D MOT benchmark
-
Download the 2D MOT benchmark (data and development kit) from https://motchallenge.net/data/2D_MOT_2015/
-
Set the path of the MOT dataset in global.m
-
Run compile.m. OpenCV is needed.
-
For validataion, use MOT_cross_validation.m
-
For testing, use MOT_test.m
We provide our own detection using SubCNN [1] on the 2D MOT 2015 dataset here.
Important: make sure libsvm-3.20 in the 3rd_party directory is used. Other versions of libsvm may not be compatible with the code.
Usage on the KITTI tracking dataset
-
Download the KITTI tracking benchmark (data, development kit and detections) from http://www.cvlibs.net/datasets/kitti/eval_tracking.php
-
Check out the kitti branch
git checkout kitti
-
Set the path of the KITTI tracking dataset in global.m
-
Run compile.m. OpenCV is needed.
-
For validataion, use KITTI_cross_validation.m
-
For testing, use KITTI_test.m
We provide our own detection using SubCNN [1] on the KITTI tracking dataset here.
References
[1] Y. Xiang, W. Choi, Y. Lin and S. Savarese. Subcategory-aware Convolutional Neural Networks for Object Proposals and Detection. In IEEE Winter Conference on Applications of Computer Vision (WACV), 2017.
Contact
If you find any bug or issue of the software, please contact yuxiang at umich dot edu