RoPE demonstrates impressive extrapolation performance in Vision Transformers, leading to improved backbone performance.