Aran Komatsuzaki   @arankomatsuzaki   6/9/2021       

Scaling Vision Transformers Train ViT with 2B parameters, which attains a new SotA on ImageNet of 90.45% top-1 accuracy. Also, achieves 84.86% top-1 acc. on ImageNet with only 10 examples per class. https://t.co/dQsrZiS05X

 Reply  0     Retweet   75      Like   383





Posted by Aran Komatsuzaki