Old Is Gold: Why Photogrammetry Still Matters in the Age of NeRF

Radiance fields won the demo. Photogrammetry still owns the deliverable.

Press enter or click to view image in full size

Radiance fields won the demo. Photogrammetry still owns the deliverable.

The uncomfortable question every reality-capture team is now asked

Somewhere in the last two years, a version of this meeting started happening everywhere.

Someone on the team — usually the most enthusiastic person in the room — opens a laptop and flies through a Gaussian splat of a job site. It’s gorgeous. Reflections behave. Foliage doesn’t dissolve into cotton wool. Thin railings actually look like railings. The whole thing trained in twenty minutes on a single GPU.

Then someone senior asks the question that ends the fun: “Great. Can we measure off it?”

That question is the entire subject of this article. Because the honest answer is more interesting than either camp wants to admit. It’s not “no, radiance fields are toys.” It’s not “yes, photogrammetry is obsolete.” It’s: these two families of methods optimize for different things, and the thing you optimize for determines what you’re allowed to do with the output.

That distinction is not a technicality. It’s the difference between an asset you can put in a design review and an asset…