Fast PolyReduce SOP helps you to reduce the polycount of high resolution geometries in a very fast way, thanks to parallel processing.
Feedback and testing: You are very welcome to report here any issue and feature request. Documentation is limited, but there is an handy Examples File that showcases some comparison and strengths of the tool. Also, PLEASE somebody could test it in a real word scenario and report back (like with sculpts or 3D scans, 10+ millions polys)? I have only tested it with "dummy" heavy geometries which could not represent real world usage.
Up to 22.5x faster than traditional PolyReduce SOP (tested on 24 cores machine)
Possibly better quality topology than traditional PolyReduce SOP when source geometry is quads-based and you need to generate a low-res triangles-based mesh.
Quads-only output with Instant Mesh wrapper
Up to 8x speed faster than vanilla Instant Mesh
Best speed gain with very HIGH-RES geometries and HIGH-RATIO of polycount reduction.
Not as many features and controls as the traditional PolyReduce SOP
There is an automatic cap at 50%: you can generate only an amount of polys lower than half of the input polycount. Meaning that if the input geometry has 100000 polys, the allowed output polycount is capped at 50000 polys. You can't reduce a 100k polys geo down to a 52k polys. This is generally not an issue because you normally want at least to halve the high-res input.
When generating quads, attributes and UVs are destroyed. This is a limitation of Instant Mesh SOP.
How it works?
At its heart there's the previously released Fast Remesh SOP [www.sidefx.com], which splits the input geometry in many clusters and remesh each one inside a compiled for-loop. Se we are talking parallel processing. This is where the enormous speed gain come from.
At this point the geometry is so much reduced in the polycount, that the traditional PolyReduce Sop can finally take over and work on the messy topology produced by Fast Remesh to output a clean result at high speed.
As I learn to work with this new tool, I noticed the following:
Performance curve of Fast PolyReduce is opposite than in the traditional Polyreduce, meaning that: as you perform more aggressive reductions, you get a speed boost in Fast PolyReduce, but you get a speed penalty in the traditional one.
Of course, the latter can take advantage of its "smart caching" feature, so that it perform much of the computation only the first time it sees a new mesh, then you can have subsequent reductions almost for free, if you decide to try different reduction ratios. However it seems that on very heavy geometries Fast PolyReduce can still contests the performance of the traditional PolyReduce in the second and following reductions.
But if we just speak about first runs: with the traditional Polyreduce, a first run of reduction down to 10% of the original polycount takes more time than a first run set at 50%, which takes more time than a first run at 90%. The opposite happen with Fast PolyReduce. So the best scenario for this new tool is when you want a high ratio of polycount reduction from very heavy geometries.
Regarding this smart caching feature of the traditional PolyReduce, it might possible to have it as well in Fast Polyreduce, since it's using Polyreduce SOP inside, why not taking advantage of it?
Fast PolyReduce seems to be very promising for 3D scans. In fact I see fewer topology artifacts when the source has a more organic and detailed topology than when it has clean shape... These artifacts are the seams of the clusters used by the inner Fast Remesh for parallel remeshing. In some cases they are leftover in the final poly reduced output, especially when using a Base quality preset.
Hello Dalorin, the tool is already free to use in commercial projects, but it was made with Houdini Indie. I realize now that owners of FX, CORE and studio versions of Houdini are not allowed to use simple HDALCs?
I'll see what I can do, maybe re-publishing Fast PolyReduce through Orbolt which should convert it to a full commercial license.
1- Some adaptivity controls exposed: now you can set equalize length and use a custom attribute to control the reduction (like in the vanilla Polyreduce)
2- No more quads output. The Instant Mesh Sop wrapper is now gone in order to make the asset compatible with Orbolt (at this point no clearance to use this Labs node in Orbolt. It might change in the future) Of course you can just manually append the Instant Mesh Sop after Fast PolyReduce and you will get the very same results and speed increase as in the previous version.