When_Vision_Transformers_Outperform_ResNets_without_Pre-training_or_Strong_Data_Augmentations