Bolt example STL non-manifold

[lynaghk]

First, let me say thanks for making this cool library and sharing it! I'm thrilled to see SDFs of so many kinds of screw threads, they're much more interesting to me than that sphere subtracted from a cube nonsense every other SDF library has as the headline image =D

Anyway, I ran with go version go1.24.2 darwin/arm64 on an M1 Macbook Air with MacOS 14.7.5 (23H527) and ran into the same issue as #8.
I really wanted an STL, though, so I modified the bolt example with VisualOutput: nil here:

	err = gsdfaux.RenderShader3D(object, gsdfaux.RenderConfig{
		STLOutput:    fpstl,
		VisualOutput: nil,
		Resolution:   float32(resolution),
		UseGPU:       useGPU,
	})

Unfortunately, the STL looks much worse than the one you got in your README. I'm rendering with OrcaSlicer (background) and F3D (foreground).

Just figured I'd let you know. Have a great day!

[joeblew999]

Glad to see it’s not just me hitting Mac compile issues .

I am also running on windows arm64 .

[soypat]

A few things to note:

• The images shown in the readme are SDF visualization. You are straight up visualizing the raw mathematical field using raymarching, so you'll get perfect quality visualization of the shape
• The STL is built with triangles using a rendering technique. The default algorithm is Marching Cubes algorithm.
  • Marching cubes will yield choppy quality on the threads unless you reduce the resolution.
  • You can try the DualContour renderer which in theory should yield much higher quality triangle mesh, but I still have not gotten it to work as per the reference https://github.com/cheind/sdftoolbox
  • @joeblew999 Totally missed your issue on compile issues on Mac. Will take a look

Any help with getting the DualContour to actually work like the reference is greatly appreciated

[lynaghk]

Gotcha, thanks for the details. Makes sense the output is different if there are two different algorithms in play (SDF to pixels vs. SDF to triangle mesh).

My background is more from the NURBS/B-Rep CAD side of things, so I won't be able to help with the graphics stuff. I will pass it along to my SDF crew though and see if any of them might be interested in having a go.

Also, I have no idea if it's relevant or helpful, but Manifold has, according to their README,

a level set function for evaluating signed-distance functions (SDF) that improves significantly over Marching Cubes.

so that might be worth investigating as well for anyone interested in this quest.

[joeblew999]

thanks @soypat I will take a look at the branch and try it on Mac, Windows and Linux.

If you are on a Mac, UTM can do all this for you too btw !!

[soypat]

FYI: I've improved the Dual Contourer implementation in #15

Images in the PR to show perfromance for a snowman example. Still does not capture edges perfectly but at least it now does not artifact terribly.