Joint Image Training for Transformers in Supervised Learning

A regularization method for ViT. Works surprisingly well on small scale datasets!