Weak result when using my own video

[ahmadi3d]

Hello I just downloaded a traffic video from YouTube : https://www.youtube.com/watch?v=mrjqS5zd268&ab_channel=TaverNet I gave these values as ground truth : 258.0000,698.0000,388.0000,201.0000 When I run the tracker it starts great but as soon as a very small occultation happens it will lose the target and keeps changing targets! if you try the sample you can see what I mean Reading the paper I understood the process and it shouldn't be so weak against occultation and disappearances, I don't understand why something like this is happening here, maybe its a bug ?, I would really appreciate it if you could check it out and tell me what I can do to fix this, thank you.

[pvoigtlaender]

I think this sequence is just extremely hard due to the high object density. You can try to tweak the hyper-parameters, especially the weighting factors of the TDPA.