Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Centroid-based association in OCSORT keeps switching IDs for the same detection #1308

Closed
1 task done
ThenoobMario opened this issue Feb 27, 2024 · 5 comments
Closed
1 task done
Labels
question Further information is requested

Comments

@ThenoobMario
Copy link

Search before asking

  • I have searched the Yolo Tracking issues and found no similar bug report.

Question

Hello, I am experimenting with the centroid-based association for my use case where I won't be able to process a lot of frames at the same time. I went through #1246 where it showed good results. However, when there are two persons in the frame, the id keeps on switching between the two, alternatively showing 1 and 2 over the same person.

What can be the issue?

@ThenoobMario ThenoobMario added the question Further information is requested label Feb 27, 2024
@mikel-brostrom
Copy link
Owner

If the threshold for the association is too large. There is certainly a risk of objects being mixed up. Try a lower threshold.

@ThenoobMario
Copy link
Author

Hi @mikel-brostrom, thanks for the reply. I am attaching my configurations.

tracker = OCSORT(det_thresh = 0.4,
                 max_age = 120,
                 min_hits = 3,
                 asso_threshold = 0.2,
                 delta_t = 3,
                 asso_func = "centroid",
                 use_byte = False)

The ID switching is still an issue. I tried keeping the asso_threshold high as well as low.

@mikel-brostrom
Copy link
Owner

Centroid-based association may be problematic if both objects are within each others association distances. There is no magic solution for this...

@ThenoobMario
Copy link
Author

I do understand it can be problematic for objects that are within each other's association distances. Here is a case. Will this count as objects being within their association distances?

@mikel-brostrom
Copy link
Owner

The performance of your object detector seems to be very poor. A lot of flickering is noticeable in the video. This is clearly not a problem mainly of the tracker but of the object detector.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
question Further information is requested
Projects
None yet
Development

No branches or pull requests

2 participants