BEiT: BERT Pre-Training of Image Transformers

[jamesp]

The paper is here. Seems like an interesting approach to produce models for tasks like segmentation.