Looking for MH developer for Valentina 3D

[learning]

A good 3d cloth, suitable to be nicely deformed in Blender (or any other 3d package) need to be more "distributed". I mean more uniform, more homogeneous and more dense.

As a mockup demonstrating this... Here's how it currently looks:



Here is how it should look in order to be efficient when wrapping around a body (I only did one of the pieces):



I.e, a reasonably homogenous distribution of quadratic faces covering the entire surface.

I'm probably clueless, but I don't see what's the problem with this. I mean, aren't those pieces, basically, just 2d polygons? If so, one can just merge all the triangles into a single face, and then apply subdivisions to the face as many times as necessary until it's uniformly quadrungulated and dense enough to apply transformations to it. And then those pieces can be fitted onto the manequen (or MakeClothes base mesh for that matter) and sown together. Is there some part I miss?

[joepal]

I'm probably clueless, but I don't see what's the problem with this. I mean, aren't those pieces, basically, just 2d polygons? If so, one can just merge all the triangles into a single face, and then apply subdivisions to the face as many times as necessary until it's uniformly quadrungulated and dense enough to apply transformations to it. And then those pieces can be fitted onto the manequen (or MakeClothes base mesh for that matter) and sown together. Is there some part I miss?

There's a difference between what's possible and what gives a good result. Here's a quick demo of what happens in blender:



First I created a polygon with a single face. Then I triangulated it (using blender's automatic triangulation). Then I subdivided the edges three times.

As you can see the mesh becomes very heterogenous. In some places it's very sparse, in some it's very dense.

So the general comment is that while it is indeed possible to use these automatic tools in the way you describe, the result becomes somewhat disappointing.

[learning]

Well, I was not thinking triangulation → subdivision, but rather triangulation → quadrungulation (as in http://www.cs.mcgill.ca/~cs507/projects ... lcome.html // http://cgm.cs.mcgill.ca/~godfried/publi ... quad.ps.gz ; turns out it is always possible to quadrungulate a 2d triangulated n-gon by adding no more than floor(n/3) new points; it's strange that this algorithm doesn't seem to be implemented in blender, but this can be fixed) → running some sort of an iterative algorithm that will split the largest faces until all of them are the same area up to a given relative threhold.

[joepal]

Sounds like a likely plan. My main point with the first set of illustrations was merely that it's desirable to have a homogenous mesh covering the surface, and that blender needs some sort of help with that. Be it inside blender or in some kind of export step before blender. It's probably possible to script it inside blender once one knows how to calculate it.

[learning]

I think the main problem is quadrungulation, because MH only works with quads. As soon as this is achieved, everything else is more or less known algorithms. Man, I can totally see in my head how it should work! My knowledge of python and Blender API's in particular is, however, lacking…

[joepal]

Why bother with it then? The first step is converting a valentina pattern to a reasonably functional wavefront obj. You don't necessarily have to do that with neither python nor blender. Wavefront obj isn't a particularly complex format either.

[brkurt]

I think the main problem is quadrungulation, because MH only works with quads. As soon as this is achieved, everything else is more or less known algorithms. Man, I can totally see in my head how it should work! My knowledge of python and Blender API's in particular is, however, lacking…

Having beta-tested Marvelous Designer 4 for three months, I would suggest that Valentina code be set up such that Valentina *only* creates quads--right from the start--and nothing else; MD4's attempt at pure-all quad solutions has not worked for them, as you can see from Thomas' version of the software.

[learning]

Having beta-tested Marvelous Designer 4 for three months, I would suggest that Valentina code be set up such that Valentina *only* creates quads--right from the start--and nothing else; MD4's attempt at pure-all quad solutions has not worked for them, as you can see from Thomas' version of the software.

Then probably the 2d polygon quadrungulation algorithm needs to be included as part of Valentina's .obj export routine. As far as I'm concerned, that's same difference. But then it's only the current devs who can put it in, as its code documentation is like this: http://valentinaproject.bitbucket.org/d ... index.html

[ThomasL]

Quads is not the only requirement. As I pointed out somewhere above, you can always quadrangulate by subdividing a triangle mesh, but there are fatal drawbacks:

1. Bad topology.
2. Too dense mesh.
3. Too high valency.

[learning]

Yeah, those are valid concerns. Especially the valency problem — what if you triangulate and get a fan-like topology? The algorithm in the paper quoted above is not guaranteed to be able to handle these. It's a bit simpler in 2d, but still needs some thought. The approach I was thinking about some time ago is based on "cutting out rectangles" — i.e. first subdividing away the largest rectangular shapes one can fit into a given polygon and then dealing with what remains…