Twins: Revisiting the Design of Spatial Attention in Vision Transformers Very recently, a variety of vision transformer architectures for dense predic