Home

Awesome

A Library of Multi-Object Tracking in Python and Pytorch

Installation

environments: python 3.6.10, opencv 4.1.1, pytorch 1.3+

git clone https://github.com/nightmaredimple/libmot --recursive
cd libmot/
python setup.py install develop --user

The details can be seen from my blogs or zhihu.

Feature Lists

BlockMethodReferenceComplete
IOU Assignmentiou-tracker&V-IOU
Linear Assignment-
Data AssociationMinCostFlowMCF
Other End-to-End NetworkDAN&DeepMOT
GNN&GCNMPNTrack
-----------------------------------------------------------------------------------
Kalman FilterSort&DeepSort
MotionECCTracktor++
Epipolar GeometryTNT
-----------------------------------------------------------------------------------
Re-ID-
AppearanceFeature Fusion&Selection-
DANDAN
-----------------------------------------------------------------------------------
DetectionFaster RCNN + FPNTracktor++
-----------------------------------------------------------------------------------
SOTCF&SiamKCF&CN
-----------------------------------------------------------------------------------
DataLoader-
TricksSpatial Blocking-
-----------------------------------------------------------------------------------
Evaluation-
OthersTracking Visualiztion-
Feature Visualiztion-
-----------------------------------------------------------------------------------
TracktorMIFT(ours)-
-----------------------------------------------------------------------------------
DetectorMIFD(ours)-

Motion Model

python scripts/test_kalman_tracker.py
<div align="center"> <img src="figures/kalman_tracker.png" /> </div>

Data Association

<div align="center"> <img src="figures/linear_assignment.png" /> </div>

Tracktor

Our proposed MIFT and MIFD will be released upon acceptance.

In MOT Challenge, the MIFT tracktor is named as ISE-MOT, the MIFD detector is named as ISE-MOTDet.

MethodDataSetsMOTA↑IDF1↑MT↑ML↓FP↓FN↓ID Sw.↓Frag↓Hz↑
MOT1548.152.129.5%26.2%102462084077611976.7
MIFTMOT1660.457.324.6%28.9%5510667237049326.9
MOT1760.156.228.1%27.8%22265200077264432067.2
MethodDataSetsAP↑MODA↑FAF↓Precision↑Recall↑
MIFDMOT17Det0.8870.74.181.491.7

To be continued..