Tracking seems to rely too much on test results

[22xxlin]

Search before asking

• I have searched the Yolo Tracking issues and found no similar bug report.

Question

Hello, I have a question to ask
My understanding is that tracking can be aided by trajectories when no object is detected, for example
For example, in Botsort, track_buffer:
type: randint
default: 81 # from the default parameters
range: [20, 81]

In my scenario, the object was not detected for about 70 frames in the middle of the video input, but it was OK before and after that. I thought tracking would help to some extent, but I found that adjusting the track_buffer value made no difference in effect

My target object, from the original tracking ID1, and then in the middle of more than 70 frames without tracking and detection results (because the target and the background are indeed too similar, the detection is very difficult), the object behind became the new tracking ID2.
If you need more detailed video effects I can provide, but I would like to ask why