Vision Transformer / ViT (2020)

Dosovitskiy et al. (2020) applied transformers directly to image patches, proving that pure attention without convolutions matches or beats CNNs on image classification.