Why here is negative MCM score？

[LuFan31]

Hi, thank you for your excellent work. But I have a question when reproducing the code. Why is a negative MCM score used in Line.248 in MCM/utils/detection_util.py? According to convention, ID samples should have higher scores, and ID samples obtain higher multimodal similarity in the CLIP model. Therefore, shouldn't we not use the negative sign here?

[alvinmingsf]

Thanks for your interest in our work! This is because in L259 in detection_util.py, the function takes in -in_score, -out_score as inputs. So if you implement in a way that removes negative signs in both places, it should also work.

aurocs, auprs, fprs = [], [], []
measures = get_measures(-in_score, -out_score)

[LuFan31]

Thanks for your apply!