ID shifting

[Ratnadeep22]

Is there any way or parameter tweaking to solve the issue of ID SHIFTING. I have tried the following parameters:

1. nms_max_overlap
2. max_age
3. n_init
4. NearestNeighborDistanceMetric parameters.

Is there any other way or I am missing any parameter which can solve this issue. OR I suppose to train deep-sort for the scenario where I am using it?

---

Thanks and regards,

[nwojke]

What is ID shifting exactly?

[Ratnadeep22]

@nwojke thanks for your quick response. Id shifting means when two semi occluded pedestrians walk together having tracker id's 1 and 2 resp., after some frames, their ids get exchanged.

Thank you!

[cesarandreslopez]

Same question here. Any suggestions would be welcomed. @nwojke @deamonDevs

[lg-code-repo]

it seems to be IDSwitch

[SuTanTank]

ID switching is well-controlled for the MOT test videos. But when I tried with other videos with my own detection results (in MOT 10-column format), ID switching is everywhere, even in the same MOT videos with different detectors (I used YOLOv3). The detected bounding boxes seem good enough to me, so I'm confused why just slightly different detection can cause such huge ID switching issue.

[zhouzw87]

Same question here, and if my detection-module not detected object, even only one frame, but next frame detected object, ID changed the new ID. the occlusion also is the same situation, one object was occluded, When it appear again, it'sID will chang.

[duyanfang123]

How to solve this problem?

[anzy0621]

Has anyone found a solution for this?

[Harishyadav766]

i am also facing the same issue, my id's are switching. If anybody have solution for these please let us know here (I used YOLOv3). The detected bounding boxes seem good enough to me, so I'm confused why just slightly different detection can cause such huge ID switching issue.

[luvwinnie]

any progress on this? I have the same problem

[Harishyadav766]

nope.

On Fri, Sep 4, 2020 at 6:59 PM Leow Chee Siang [email protected] wrote:

any progress on this? I have the same problem

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/nwojke/deep_sort/issues/88#issuecomment-687145871, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANCYYRXJL4SUU7YATMPEOPDSEDTTHANCNFSM4FJ2Z4NA .

--

Pyadindi Harish Yadav M.Tech

Junior Developer

M: +91 8639581806 , +91 7799059796

E: [email protected] [email protected] - www.kayadev.ai

A: #34, 6th Main, Banaswadi Main Road, Bangalore, 560043

[image: Facebook icon] https://www.facebook.com/kaya.dev.solutions/ [image: LinkedIn icon] https://in.linkedin.com/company/kayadev

[Shrustishree]

Any improvement for this ID switching? I am using yolov4 deepsort model.

[Durys]

Hi - how is work going to solve this issue? I was thinking of tweaking the hyper=-parameters on the deepsort model in order to minimize the occurence of such problem. Maybe a grid search would solve this issue?

Cheers!

[zhouzw87]

Hi - how is work going to solve this issue? I was thinking of tweaking the hyper=-parameters on the deepsort model in order to minimize the occurence of such problem. Maybe a grid search would solve this issue?

Cheers!

Maybe, Replace existing detection algorithms with anchor-free algorithms, which should alleviate the problem..

[Durys]

Hi - how is work going to solve this issue? I was thinking of tweaking the hyper=-parameters on the deepsort model in order to minimize the occurence of such problem. Maybe a grid search would solve this issue? Cheers!

Maybe, Replace existing detection algorithms with anchor-free algorithms, which should alleviate the problem..

How would anchor-free algorithms help here?