Awesome
Lifted Disjoint Paths with Application in Multiple Object Tracking
This is the official implementation of our ICML 2020 paper Lifted Disjoint Paths with Application in Multiple Object Tracking (Andrea Hornakova, Roberto Henschel, Bodo Rosenhahn, Paul Swoboda) [https://arxiv.org/abs/2006.14550].
We provide the solver implemented in C++, along with a Python wrapper. The tracker is implemented in Matlab.
Evaluation
Using the features explained in our paper, we achieve the following results on MOT17:
MOTA | IDF1 | FP | FN | IDs | |
---|---|---|---|---|---|
Train | 67.0 | 72.4 | 2655 | 107803 | 791 |
Test | 60.5 | 65.6 | 14966 | 206619 | 1189 |
Note that all results on the training set have been calculated in a leave-one-out fashion so that values are actually meaningful.
Citation
If you use our work in your research, please cite our publication:
@InProceedings{lifted_disjoint_paths_2020_ICML,
author={Andrea Hornakova and Roberto Henschel and Bodo Rosenhahn and Paul Swoboda},
title={Lifted Disjoint Paths with Application in Multiple Object Tracking},
booktitle = {The 37th International Conference on Machine Learning (ICML)},
month = {July},
year = {2020}
}