Yolov3 (for different classes) + GOTURN

[MuhammadAsadJaved]

Hi,
Thanks for this great work. I have a question. I am trying Yolov3(detecting person and car) +DeepSort in Xavier NX. But the speed is very slow. Maximum 10 FPS with TRT optimization.

So can I combine the GOTURN tracker with my YOLOV3 for this task? Can this also track person and car or I need to train tracker as well?
Does the more number of classes , i.e 6 classes will affect the speed too much?

[Boinet]

Hi @MuhammadAsadJaved,
Thanks for your interest.
Indeed this project combining GOturn ans yolo seems to be a good fit for your application.
What you need to do:

• train yolo v3 on pedestrians or whatever object you wish to detect and track. (If you already trained yolo you do not need to retrain it)
• for multiple targets, and classes, the code would need a little adaptation, but I don’t think it’ll be too much trouble.
• use our code

Fyi: the goturn is a general purpose network, meaning that it is trained to track every kind of objects. Hence you won’t need to retrain it.

Good luck,

[MuhammadAsadJaved]

@Boinet OK. Thank you so much.

[GOBish]

@MuhammadAsadJaved if you could share your code and process if you complete this successfully it would be greatly appreciated.

[MuhammadAsadJaved]

@GOBish OK. Sure.

[MuhammadAsadJaved]

@GOBish Please read issue #7 , I have explained the required libraries and installation method in the last comment.

[GOBish]

Thanks! Do you know if OpenCV versions 4+ would work?

[MuhammadAsadJaved]

It should be fine. You can have a try.

…

On Fri, Sep 25, 2020 at 7:40 PM GOBish ***@***.***> wrote: Thanks! Do you know if OpenCV versions 4+ would work? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AG4GR5FRU3ZTE4RDSZDC4WTSHR6S7ANCNFSM4RUBG5JA> .

[MuhammadAsadJaved]

@GOBish Is there any update?

[GOBish]

not yet, will update you as soon as I get up to that step.

[MuhammadAsadJaved]

OK. Thank you