If you haven't read our latest ImageNet SOTA work "Vision Transformers (ViT)" yet, shame on you. But! There's hope! Here's the corresponding blogpost which is a nice tl;dr: ai.googleblog.com/2020/12/tr…