nwojke
Where is lambda in Equation 5
I'm trying to adjust the weights between the Karman filter results (location) and the features (appearance). Can anyone please kindly help?
I would be interested in the same. Did they maybe set it to 0, and hence it is not included?
Hi, sorry for the late reply. The question has come up a few times already. Unfortunately, our final implementation doesn't contain any lambda at all (as we set it to 0). Instead, we have a function that invalidates the appearance-based cost according to the Kalman filter gate here.
I hope this helps.
@nwojke. Thanks for your comments, they are much appreciated. I have a couple of questions regarding the current implementation of the cost matrix: Are there any remarkable reasons why you did not implemented the cost matrix computation as explained in Eq.5? Is the logic behind the current implementation better? Thank you!
Hi @nwojke , I read some issues about Lambda and your clarification that you have assumed lambda as 0 and implemented the gating metric. But I have a doubt: you calculate the cost matrix using cosine distance, then you pass that cost_matrix to gate_cost_matrix which thresholds it based on mahalanobis threshold and then it is thresholded using cosine_max_threshold. (I hope I have understood the code in the right way)
Why would you need to do Mahalanobis thresholding when the cost matrix calculation is purely based on cosine distance?
I hope my question is not too confusing!
@pratikbhave2 this way the kalman-based Mahalanobis metric is still being used as a kind of mask to filter out wildly improbable locations.
I implemented it in my fork of the repository (can be found here: https://github.com/michael-camilleri/deep_sort)
@michael-camilleri Thank you for your code, but I have been trying with changing Lambda from 0 to 1 and so effect on the ID was seen , any idea why that could be ? To implement your code I only replaced the Tracker.py by your own Tracker.py file
Hi
I'm sorry for the delay in replying but it has been a hectic month.
I would recomend using the entire repository rather than replacing the tracker.py, since the differences come also in some other methods.
M
From: kameel2311 @.> Sent: 21 April 2021 09:54 To: nwojke/deep_sort @.> Cc: CAMILLERI Michael @.>; Mention @.> Subject: Re: [nwojke/deep_sort] Where is lambda in Equation 5 (#112)
This email was sent to you by someone outside the University. You should only click on links or attachments if you are certain that the email is genuine and the content is safe.
@michael-camillerihttps://github.com/michael-camilleri Thank you for your code, but I have been trying with changing Lambda from 0 to 1 and so effect on the ID was seen , any idea why that could be ? To implement your code I only replaced the Tracker.py by your own Tracker.py file
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/nwojke/deep_sort/issues/112#issuecomment-823898051, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AIKRP36SPFFFAIL3B6OAF5TTJ2HFHANCNFSM4GAUYU3A.
The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. Is e buidheann carthannais a th’ ann an Oilthigh Dhùn Èideann, clàraichte an Alba, àireamh clàraidh SC005336.