FFT(Flow-Fuse Tracker)[1] is an end-to-end DNN tracking approach, that jointly learns both target motions and associations for MOT(multiple object tracking). There are some details of reading and implementing it.

Contents

• Paper & Code & note
• Paper understanding
• Code understanding
• Note
• References

Paper & Code & note

Paper: Multiple Object Tracking by Flowing and Fusing(arXiv 2020 paper)
Code: [Pytorch][Updating]
Note: FFT

Paper

Abstract

1. Previous: estimating target-wise motions and conducting pair-wise Re-Identification(Re-ID).
2. This paper: target flowing and target fusing.
3. Achievment: SOTA on 2DMOT15, MOT16 and MOT17.

Problem Description

Reccent approaches: First produce target motion and appearance features respectively, then conduct target association between frames.
It is very difficult for computating both features.

Problem Solution

It includes two techniques: target flowing and target fusing.
For target flowing: FlowTracker extract target-wise motions from pixel-level optical flows.
For target fusing: FuseTracker refines and fuse targets.

Conceptual Understanding

Experiments

Code

Algorithm

Note

References

[1] Zhang, Jimuyang, et al. “Multiple Object Tracking by Flowing and Fusing.” arXiv preprint arXiv:2001.11180 (2020).