About the model Resnet50 about moon

[QinbinLi]

Hi @Skye601 ,

models.resnet50() is the official implementation of torchvision while ResNet50_cifar10() is a better implementation for cifar-10 with smaller kernel size.

[Skye601]

Thanks for your reply. when training on tinyimagenet, should models.resnet50() or ResNet50_cifar10() be used? If there exists the model head ( two FC layers), it seems like choosing ResNet50_cifar10().

[QinbinLi]

Hi @Skye601 ,

I used ResNet50_cifar10() in the paper's experiments. You can also try models.resnet50() to see which one has a better performance.