The question of IRM input

[ricky-696]

Hello~ Thank you for your amazing paper.

I would like to ask about the input of IRM.

In Figure 3, the input is described as $X^t=(F^t, \hat{B^t})$. but in Equation (1), we can see that the Removal branch only takes $2F^t$ as input without passing through self-attention.

Could you clarify whether the input for self-attention in both branches includes the complete $(F^t, \hat{B^t})$ as $X^t$ , or if the Removal branch only takes $F^t$ as input, while the Addition branch only takes $\hat{B^t}$ as input?

If you could also provide the corresponding location of this input in the code, that would be even better!

Thank you.