Simplification issues using the new meshopt_generateShadowIndexBuffer

[jtorresfabra]

Hi all,

I've been working with meshoptimizer for a while, for simplification I was used to first get rid of duplicate vertices (by having just positions into account) and then simplify. I don't really care about normals or UVs in my usecase, so simplification really worked pretty well for that case reducing to 20% of the size.
The process was like:

meshopt_generateVertexRemap
meshopt_remapIndexBuffer
meshopt_remapVertexBuffer 
meshopt_simplify

This week I learned about meshopt_generateShadowIndexBuffer and thought it could work as a replacement and tried

meshopt_generateShadowIndexBuffer
meshopt_simplify

But the simplifier was not able to really work more than reducing to 97% of the size. I understand that the shadow index buffer is doing the vertex weld, so I was expecting it to work in the same way plus I wouldn't need to reorder/remap the vertices (cause I need to keep the order in that phase for a different matter). But simplification is definitely not working well.

Am I making sense or totally missing the point?
Thanks in advance.

[zeux]

That is odd. I believe these should be roughly equivalent indeed. Maybe there's a bug in meshopt_generateShadowIndexBuffer...

Could you show the code you are using before (with generateVertexRemap) and after (with generateShadowIndexBuffer)?

[jtorresfabra]

Just for the sake of completion I added a counter in meshopt_generateShadowIndexBuffer and increased it whenever if (*entry == ~0u) and stated that number of vertices matches with the uniqueVertices generated in the case of the remap. So my feelings were more that the error could be in the simplifier, because meshopt_generateShadowIndexBuffer seems straigthforward to follow. However I don't want to cause noise in your thinking approach 😅

[zeux]

Interesting. Thanks for checking the shadow index buffer code as well, I was going to suggest the counter next :) Can you also enable tracing (define TRACE to 1 at the top of simplifier.cpp, line 10) and share the output with both approaches?

[jtorresfabra]

With remap

locked: many open edges 0, disconnected seam 0, many seam edges 0, many wedges 0
position remap: 1490 vertices => 1490 positions
kinds: manifold 1187, border 303, seam 0, complex 0, locked 0
pass 0: removed 796 triangles, error 1.432280e-03 (goal 1.224227e-03); evaluated 1343/4163 collapses (done 446, skipped 855, invalid 41)
pass 1: removed 517 triangles, error 3.144471e-03 (goal 2.603031e-03); evaluated 764/2918 collapses (done 299, skipped 456, invalid 8)
pass 2: removed 597 triangles, error 5.088787e-03 (goal 4.392267e-03); evaluated 476/2098 collapses (done 322, skipped 141, invalid 12)
pass 3: removed 220 triangles, error 7.644158e-03 (goal 6.487423e-03); evaluated 241/1179 collapses (done 115, skipped 120, invalid 5)
pass 4: removed 9 triangles, error 7.644158e-03 (goal 3.961690e-03); evaluated 6/829 collapses (done 5, skipped 0, invalid 0)
result: 1602 triangles, error: 7.644158e-03; total 5 passes
Index count before: 8025. Index count after: 1602 (19.9626%). Error: 0.00764416

With shadowIndexBuffer

locked: many open edges 0, disconnected seam 0, many seam edges 671, many wedges 604
position remap: 3373 vertices => 1490 positions
kinds: manifold 48, border 167, seam 0, complex 0, locked 1275
pass 0: removed 151 triangles, error 2.683837e-03 (goal 0.000000e+00); evaluated 337/337 collapses (done 112, skipped 214, invalid 11)
pass 1: removed 21 triangles, error 4.984526e-03 (goal 0.000000e+00); evaluated 72/72 collapses (done 13, skipped 54, invalid 5)
pass 2: removed 0 triangles, error 4.984526e-03 (goal 0.000000e+00); evaluated 10/10 collapses (done 0, skipped 0, invalid 10)
result: 7509 triangles, error: 4.984526e-03; total 3 passes
Index count before: 8025. Index count after: 7509 (93.5701%). Error: 0.00498453

Seems something is happening here
many seam edges 671, many wedges 604
kinds: manifold 48, border 167, seam 0, complex 0, locked 1275

[zeux]

Ohh. Uhh. Yes :( Thanks for bringing this to my attention! This never worked and needs some sort of tweaks.

I believe you can use meshopt_SimplifySparse option which should fix the issue (for your usecase you can ignore the absolute error flag even though the documentation advises it). The documentation right now states that you can use shadow index buffers as an input and that is incorrect. I will remove the mention of this from the documentation for now and add a TODO item to investigate the best option to fix this.

[jtorresfabra]

Well, thanks a lot! I will give a try to meshopt_SimplifySparse and comeback with the results, just in case other people is interested. Thanks for your work and your quick reply!

[jtorresfabra]

Just a note to say it worked as expected, just added meshopt_SimplifySparse to the simplifier flags and giving the shadowIndexBuffer as input I get:

locked: many open edges 0, disconnected seam 0, many seam edges 0, many wedges 0
position remap: 1490 vertices => 1490 positions
kinds: manifold 1187, border 303, seam 0, complex 0, locked 0
pass 0: removed 798 triangles, error 1.432280e-03 (goal 1.211187e-03); evaluated 1341/4163 collapses (done 447, skipped 848, invalid 45)
pass 1: removed 513 triangles, error 3.144471e-03 (goal 2.603032e-03); evaluated 761/2915 collapses (done 297, skipped 453, invalid 10)
pass 2: removed 591 triangles, error 5.320754e-03 (goal 4.383419e-03); evaluated 474/2101 collapses (done 319, skipped 138, invalid 16)
pass 3: removed 220 triangles, error 7.644158e-03 (goal 6.670701e-03); evaluated 239/1191 collapses (done 115, skipped 114, invalid 9)
pass 4: removed 17 triangles, error 7.644158e-03 (goal 4.367802e-03); evaluated 15/841 collapses (done 9, skipped 2, invalid 3)
result: 534 triangles, error: 7.644158e-03; total 5 passes
Index count before: 8025. Index count after: 1602 (19.9626%). Error: 0.00764416

Thanks again!