Supplementary Material for DropPos: Pre-Training Vision Transformers by 1 Reconstructing Dropped Positions

More details and examples can be found in [24]. A.2 COCO object detection and segmentation Experiments are conducted on 8 Tesla V100 GPUs. All experiments are conducted on 8 Tesla V100 GPUs. We underline the special parameter different from our default settings. Attention is all you need.