difference about DAN and DHN

yihongXU / deepMOT

Official Implementation of How To Train Your Deep Multi-Object Tracker (CVPR2020)

GNU Lesser General Public License v3.0

498 stars 87 forks source link

difference about DAN and DHN #3

Closed wangxianrui closed 5 years ago

wangxianrui commented 5 years ago

notice that you use the DHN in training， but use DAN in evaluate， what is the difference, and how the DHN influence the tracking result ?

yihongXU commented 5 years ago

Hi, sorry about the notation issues. DHN is only for training, which acts as a proxy to deliver gradient from MOT loss. DAN is used as an appearance model, which can recover a lost track by comparing the appearances before and after the object is lost.