Well, it's quite a typical accuracy if you train on Tiny ImageNet from scratch.
If you want bigger accuracy, you need to do a lot of hyperparameter tuning. But I doubt that you can get more than ~55%.
If you want to reproduce the original paper's results, you have to train on
the standard ImageNet dataset.
Well, it's quite a typical accuracy if you train on Tiny ImageNet from scratch. If you want bigger accuracy, you need to do a lot of hyperparameter tuning. But I doubt that you can get more than ~55%.
If you want to reproduce the original paper's results, you have to train on the standard ImageNet dataset.