Hello, I saw you replying to someone else's question on the tracking page of YOLOv8.

[xxxyyyzzz3984]

Hello, I saw you replying to someone else's question on the tracking page of YOLOv8.

[xxxyyyzzz3984]

Hi, sorry for the late reply. The fact that the detection model manages to detect your target object does not necessarily guarantee the tracking model also works. The tracking model can fail to track your target object. There are many reasons leading to this result, perhaps your target object is too small or moving to fast between frames such that the almost no hit between frames can be tracked. Also may because the pretrained weights or the tracking method does not work for your situation. You can try the following see if works better for you:

Use HybridSORT algorithm instead of the default BoT-SORT or ByteTrack or other tracking methods. You can add a call_back on yolo to substitute the tracking method on on_predict_start method.

[xxxyyyzzz3984]

I used to have a scenario that requires me to track some fast moving ping pong balls. I failed to track them while using all tracking methods (e.g., BOTSORT, ByteTrack, DeepSort, etc) until I used HybridSORT. Then I successfully tracked those fast-moving small ping pongs. Hope it helps!

[karlmaji]

Thank you very much for helping me understand this issue. I tried some methods and used weights with higher detection accuracy, and now the problem has been resolved. I will also try some of the other tracking algorithms you mentioned, which should yield better results.