Question about computational complexity formulation of Outlooker Attention

[ligeng0197]

Greetings! Thanks for all your inspiring and excellent VOLO work!!! In reading this paper, I get trouble in comprehending the formulation (8), which depicts the complexity of Outlooker Attention. I tried to inference the cost from the pytorch-like code provided aforementioned, however cannot get to the formulation (8). Would you mind providing any insight about the calculation process? Thanks a lot.