Abstract: 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 ...