Using Pretrained-embeddings along with custom trained detections

[AhmedKhaled945]

So i was trying to train for tracking, using qdtrack association, and this requires a lot of computational power, which i can get, but first i wanted to test how efficient will the method be,

I have a custom detector that i trained, can i use this detector for detections, and your pretrained model for embeddings and id association, or that would tear up the association accuracy?

Thanks in advance.

[MasterBin-IIAU]

Hi, it is possible to use your own detector for detecting objects and then use our embeddings for the association. The embedding is obtained by extracting features from the centers of the objects. But to achieve this goal, you may need to rewrite some code.

[AhmedKhaled945]

sure no worries,

Also wanted to ask about something, i have this case in my dataset, that has overlapping of moving objects (occlusion),
Consider two objects of same class (buses), both are moving, detected correctly, tracked correctly, until they overlap, once they overlap and one of them occlude, it comes from the other side, redetected, but both of them swap ids, this happened while using the ByteTracker association method,

Is there any solution for this, i think kalman filter should do better since objects are moving in opposite direction, so how did an id become associated to an object moving in the opposite direction (the other bus overlapping),
Regards,

[AhmedKhaled945]

If i may ask, what is the default weights and exp file for this, i mean if i am willing to use qdtracking association, what is the best trained model in the model zoo to extract embeddings to qdtracker?

I tried unicorn_det_convnext_large_800x1280 and it didn't gave me any results, along with exp file unicorn_track_large_mot_challenge.py, are these the default that should be used?

@MasterBin-IIAU