About The Given CIFAR-100 Models

[altted]

Do the given CIFAR-100 models trained from Preact-ResNet 18? I met size mismatch problems while loading them to Preact-ResNet 18.

[YanghaoZYH]

@altted As described in our paper, it was ResNet 18 for CIFAR-100, not Preact-ResNet 18. We followed the baseline method WAT, also if I remember correctly, the dataset is processed with 01 normalization rather than std.

[altted]

@altted As described in our paper, it was ResNet 18 for CIFAR-100, not Preact-ResNet 18. We followed the baseline method WAT, also if I remember correctly, the dataset is processed with 01 normalization rather than std.

Thank you, and I want to know how to train the model in CIFAR-100. Apart from changing num_classes(10 -> 100), where should I modify?

[altted]

@altted As described in our paper, it was ResNet 18 for CIFAR-100, not Preact-ResNet 18. We followed the baseline method WAT, also if I remember correctly, the dataset is processed with 01 normalization rather than std.

Thank you, and I want to know how to train the model in CIFAR-100. Apart from changing num_classes(10 -> 100), where should I modify?

[hyukahn16]

Hi. Is there any update to this?