About 'Exploration of Appearance Control Mechanism' in paper

[burnquiet]

Thanks for your Great Work.
It's mentioned that you conduncted a expriment on self-attention, reference image and the noisy image are both forwarded through the diffusion UNet with their self-attention layers connected.
I‘m a little confused about "self-attention layers connected", it's like ReferenceNet concat on spatial dim or some other ways?