Johns Hohns
Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet (ICCV2021)