Questions of diving into the codes of the Fusion part

[Cloud65000]

Hi! I'm a new learner of groundingdino. When I dive into the codes of the Phase A Fusion of groundingdino, I realize that the code firstly do the iamge-text cross-attention, later the text self attention, then the image self attention. It seems a little bit different from the dsicriptions of the original structure of the groundingdino model in the paper, which firstly do the image deformable attention and text self attention then the cross attention from text to image and from image to text. Is that True or whether I overlook some details of the code?