CODI: Improving Self-Supervised Vision Transformers with Cross Distillation
Self-supervised Vision Transformers learn strong image-level representations, but their patch-level representations can be unreliable for dense prediction.
arXiv - cs.CV (CV)2 min read