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

Parameter Tuning #154

Open
AMYEY opened this issue Jul 6, 2021 · 3 comments
Open

Parameter Tuning #154

AMYEY opened this issue Jul 6, 2021 · 3 comments

Comments

@AMYEY
Copy link

AMYEY commented Jul 6, 2021

Thank you for your tutorial, it has helped me. However if any object is occluded in a frame and after few frames the occluded object reappears, the object ID changes. This is happening quite a lot. ID switches occur every time two persons clash. Any suggestions to fine tune the parameters? I would also like to understand the max_iou_distance and n_init parameters.

@pythonlessons
Copy link
Owner

This is the fault of the tracker, I spent around 3 weeks understanding how the tracker works, and so on... If you are ready to spend 3 weeks just to understand the tracker, go for it. But I will tell you if you want to make it work better, you need to train your own tracking model or rewrite tracker code that tracker would be able to revive dead tracks and compare new tracks against dead tracks by using model predicted features.

@AMYEY
Copy link
Author

AMYEY commented Jul 6, 2021

Thank you for fast response. I need to understand the tracker parameters, any suggestions from where I could understand these...for example the max_iou_distance. Or any resource you could suggest in understanding of how tracker works would be really helpful. I cannot find much online.

@pythonlessons
Copy link
Owner

pythonlessons commented Jul 8, 2021

I can't give you one sentence answer, if you need understanding you need to read the code. max_iou_distance is used to calculate distance between two objects in Kalman filter

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

No branches or pull requests

2 participants