Understanding training mechanism of visual prompting module

[Germany321]

I notice that the parameter $$\phi_p$$ is trained to output boxes and masks. If both the input and output are boxes, then the parameter $$\phi_p$$ is trained to learn a function f(x)=x?