How to evaluate the model performance on different Confidence Threshold?

[Yash-Ark]

The performance of my six-class model is declining as a result of a data imbalance problem since there are fewer objects being counted. Nonetheless, from what I've seen, every class does well within its assigned confidence level. Nonetheless, the activity I would like to accomplish is not achievable as YOLOX evaluates the model's performance on Global Confidence. Please let me know whether this functionality has previously been built and advise what adjustments I need do to accomplish this.