Mesh face count reduction for texturing workflow

[lipsmith0701]

Hi, first of all, thanks a lot for this great project.
The reconstruction quality of gs2mesh is really impressive — the geometric details and overall results look great.

I have a question regarding the exported mesh. In my use case, the generated mesh tends to have a relatively high face count, which makes downstream steps like texture mapping a bit challenging.

I was wondering if there is any recommended approach to reduce the mesh complexity while preserving most of the geometry quality.

[Anttwo]

Thank you for the nice words!

Yes, we do!
While our meshes are lighter than concurrent works, they can still weigh between 50 and 300MB.

To make them lighter, I recommend using QEM (Quadric Error Metrics) decimation.
However, I don't recommend using the QEM implementation from Meshlab or Open3D, as they surprisingly give very weird results with our meshes.

I recommend using the QEM decimation from Blender, as it works amazingly well.

We have a python script to run this decimation approach automatically, but I recommend doing it in Blender for more natural control. You need to add a modifier called Decimate to the mesh; Stay on the default setting called Collapse, you just need to set the ratio between 0 and 1. Here are some details:
https://docs.blender.org/manual/fr/dev/modeling/modifiers/generate/decimate.html

Here's the mesh of the blue&white robot from our website gallery (you can see it has a lot of triangles):

If you apply the QEM decimation from Blender, you can drastically reduce the nb of triangles and preserve a good mesh:

This last example below has 90% less triangles than the base mesh, and still looks pretty good (so the weight of the final mesh is 10% of the initial weight, it's only a few MB):