confidence score calculation

[FatihcanUslu]

Hi there, thanks for your great work on this repo.
i am trying to calculate confidence score based on the tracker's network output. So i created an MLP block for confidence that takes 2 input and produces an output that has range 0-1. For the inputs i used x_attn rather than attn. First input is x_attn that comes from static template(always first template) inference. Second input is x_attn that comes from dynamic template inference. So i run backbone twice for confidence. But this inputs doesnt seems to work right. Confidence head, starts to learn objects in dataset rather than differenciate them. So can you give me some advice about input, model selection or anything else that can help?
thanks :)