valencebond/Rethinking_of_PAR

abnormal results on PK100K

Opened this issue · 1 comments

when i train this code, i got higher results on train set(ma: 94.4 | acc: 93.0 | Prec: 95.7 | Rec:96.4 | 96.1
), but got lower results on test set(ma: 53.4 | acc: 45.4 | Prec: 60.6 | Rec:62.0 | 61.3)

I achieved ma: 82.05 on val set and 78.88 on test set. I think the reported result is close to my experiments( on val set). But I also curious about the gap between val set and test set