Training data-efficient image transformers & distillation through attention

                 Hugo Touvron 1 2 Matthieu Cord 1 2 Matthijs Douze 1
                Francisco Massa 1 Alexandre Sablayrolles 1 Herve Jegou 1

              Abstract
                                                 ↑
    Recently, neural networks purely based on atten-
    tion were shown to address image understanding                             
    tasks such as image classification. These high-
    performing vi ...