DeiT: Data-Efficient Image Transformer

Introduction The goal of this work is not to surpass the current state-of-the-art (SOTA) in image classification, but rather to advance the capabilities of image transformers. While Vision Transformer (ViT) introduced the idea of using transformers f...