mrchenfeng / maskcon_cvpr2023 Goto Github PK

Recently I am reproducing your work MaskCon and it is a very intresting work. But I have confuse about the hyperparameters on ImageNet32 dataset of MaskCon. And your job contains hyperparameters for all other datasets except Imagenet32. Would you please offer the best hyperparameters for ImageNet32 of all methods including MaskCon, Grafit and CoIns?

Thanks for your sharing of MaskCon.
However, as I try to applied Cars196 on your method, I found it There was a decrease in model accuracy (Epoch=200). However, I found that epoch=30/50 is correct, which the performance of epoch=50 is equal to your paper indicator. Could you tell me the reason of this impunity, or do you have this impunity in your training time?
Thanks!