DropPos: Pre-Training Vision Transformers by Reconstructing Dropped Positions

Open in new window