I'm trying to improve the results I get from PolyReduce, and one feature that seems useful is “Retain Density by View” in order to help preserve the silhouette. The problem is, I can't figure out how this actually works.
The documentation on it is incredibly vague [www.sidefx.com], and I have no idea what the input should even be, let alone how I should set it up. Literally all the documentation gives me on that is “Geometry representing viewpoints”. Are these polygons? Actual points? Do those points need to have normals assigned? Do they have a defined view direction through some means such as normals?
The documentation is useless. It doesn't even give an example of this feature in use. Hilariously, it actually has a link attached to this “geometry representing view”, but the link just goes back to the top of the very same page.
I can't even find anyone talking about this feature anywhere online through Google.

Hi,
you just need one (or more) points in the third input.
no normals needed
just create a add node and place the point where the camera would be…

If you hover over “Silhouette” or “Front-Faceing” you get some info….

regards
Olaf

Is there some guideline for how the points should be placed? I tested it out at first under the assumption that it might work exactly as you described, but when I tried using a large number of points spread around the object in a circle, it actually destroyed the shape and disregarded the silhouette completely. The expected result would have been for it to perform very little decimation, but it was the opposite.
That was actually what initially lead me to believe that my assumption on how it worked must have been wrong.

All around makes no sense at all. You should create a camera and a point at the cam location. You will see it makes sense. This is what it is for: Optimising geo based on camera position or maybe a few cam positions that represent a animated camera.

I think I see where my mistake was. I assumed it would use the view positions to find the silhouette and maintain that, so I thought that using many points would give it the most accurate grasp of the silhouette.
But it sounds like it's really meant to be used with an asset you'll only see from a particular angle, so you can decimate everything that won't be seen.