You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection