How to adjust false positive outputs?

[papago2355]

Hello.
I am currently working on plastic mold anomaly classification.
The problem I am having right now is that the anomaly detection is way too sensitive.
It is way too sensitive even to those are normally considered as normal.

Here is what I am using right now.

model: padim(resnet18)
image_size : (512,1024)
normal images : 200
Abnormal images : 300

f1-score : 0.94
Auroc : 0.69

Is there some ways to adjust false positive results?
I heard manual thresholds might be helpful but I can’t find a practical example to apply manual option.