TBF that would be a good application of AI to infer what is clutter and what is not from the visual light image of the same space, the wall behind an object can simply be inferred from the sections of that wall that aren't occluded by the clutter. It might not catch some obscure cases where clutter obscures some anomalous part of the wall, but it'll be good enough in the other 99% of cases. Multiple viewpoints would probably deal with a large number of oddities too - much how photometry can get more accurate with more viewpoints already.