linyu

> I reproduce TransFG code follow the instruction with overleap on 4 Tesla GPU V-100 entirely. I just got 91.2%(Paper: 91.7%) Besides, I add AutoAug in training stage as paper...

I have not reproduce VIT on CUB using TransFG, but I think the accuracy will a bit lower than 90.8% and 91.2%

I think you can view the issue of the official DARTS version, the same question was raised. And the author of the original paper replied that DARTS' s main validation...