Method

[on] MOTS with Embedding Mask-based Affinity Fusion in Hierarchical Data Association [MAF_HDA]
https://github.com/SonginCV/MAF_HDA

Submitted on 10 Dec. 2021 02:50 by
Young-min Song (Gwanju Institute of Science and Technology)

Running time:0.09 s
Environment:4 cores @ 4.2 Ghz (C/C++)

Method Description:
Proposed method has worked in Intel i7 CPU, 4 cores
@ 4.2 Ghz, and w/o any GPU acceleration.
Parameters:
TBD
Latex Bibtex:
@article{mafhda,
title={Multi-Object Tracking and Segmentation
with Embedding Mask-based Affinity Fusion in
Hierarchical Data Association},
author={Y. Song and Y. Yoon and K. Yoon and M.
Jeon},
journal={IEEE Access},
pages={1-1},
month=may,
year={2022}
}

Detailed Results

From all 29 test sequences, our benchmark computes the commonly used tracking metrics (adapted for the segmentation case): CLEARMOT, MT/PT/ML, identity switches, and fragmentations [1,2]. The tables below show all of these metrics.


Benchmark sMOTSA MOTSA MOTSP MODSA MODSP
CAR 77.20 % 87.70 % 88.40 % 88.90 % 90.90 %
PEDESTRIAN 65.00 % 79.60 % 82.30 % 81.10 % 94.00 %

Benchmark recall precision F1 TP FP FN FAR #objects #trajectories
CAR 90.80 % 97.90 % 94.20 % 33394 725 3366 6.50 % 47952 1420
PEDESTRIAN 82.80 % 98.00 % 89.70 % 17130 349 3567 3.10 % 22798 675

Benchmark MT PT ML IDS FRAG
CAR 82.00 % 17.30 % 0.80 % 415 706
PEDESTRIAN 57.80 % 35.90 % 6.30 % 300 520

This table as LaTeX


[1] K. Bernardin, R. Stiefelhagen: Evaluating Multiple Object Tracking Performance: The CLEAR MOT Metrics. JIVP 2008.
[2] Y. Li, C. Huang, R. Nevatia: Learning to associate: HybridBoosted multi-target tracker for crowded scene. CVPR 2009.


eXTReMe Tracker