This paper proposes various techniques that help Vision Transformer (ViT) to learn small-size datasets from scratch successfully. ViT. which applied the transformer structure to the image classification task. has outperformed convolutional neural networks. https://www.mudrsat.com/product-category/bubbler/