Two questions about reference image

[Agito555]

This is a very inspiring work, and I have a few questions I haven’t fully figured out yet—I'd like to ask for your insights.

First, when generating novel views along a novel trajectory, what should be used as the reference image? Is it the reference image from the current recorded trajectory?

Also, there is an ablation study where only the reference image is used to generate the novel view—how is this implemented?