Suggestion for improving the mesher.

Next
 From:  Crusoe the Painter (CRUSOE)
1024.1 
So after being somewhat dissatified with the tri-quad mesh output, because it can generate self-intersecting meshes, and sometimes leaves out tries resulting in tiny holes, along with being very ugly in places.

Now, Zbrush 3.0 supports importing objs with n-gons, dividing them into quad-tris upon import. So I tried exporting my model obj in quad+n-gon, and importing it into zbrush. Wow! wonderful mesh, the ngons get sliced up nicely into tris!!

So my question is, why can't MOI do this? It seems the tri+quad and tri+ngon mesher are 2 separate meshers. The ngon mesher generates a much more evenly subdivided mesh. So why not junk the quad+tri mesher, and use the quad-ngon mesher to generate the mesh, and then subdivide the ngons into tris/quads? Thus avoiding the whole issue of ugly super-stretched or missing tris?

Because, this fixed up a LOT of issues I was having with really ugly thin tris, especially along the edges where 2 surfaces meet.

I'll try and post some comparisons
  Reply Reply More Options
Post Options
Reply as PM Reply as PM
Print Print
Mark as unread Mark as unread
Relationship Relationship
IP Logged

Previous
Next
 From:  Michael Gibson
1024.2 In reply to 1024.1 
Hi Crusoe,

> So why not junk the quad+tri mesher, and use the quad-ngon mesher to
> generate the mesh, and then subdivide the ngons into tris/quads?

That's exactly how it already works... If you choose quads + triangles, it starts by generating n-gons and then takes each n-gon and breaks that into triangles.

But I just checked something and it looks like I may be optimizing for triangle shape in UV space instead of 3D space, if you have surfaces that are rather elongated in one direction this may be causing some problems.

If you have a good test surface that currently generates skinny triangles, please send it to me and I will experiment with some tweaks. If you can narrow it down to just one single surface piece that is a good representative of the problem, that would help.

Also it would be great to see some sample pictures of a comparison to zbrush's triangulator so I could get an idea of what it is doing that works better.

- Michael
  Reply Reply More Options
Post Options
Reply as PM Reply as PM
Print Print
Mark as unread Mark as unread
Relationship Relationship
IP Logged

Previous
Next
 From:  Crusoe the Painter (CRUSOE)
1024.3 In reply to 1024.2 
Well, it definitely doesn't look as good as when I let Zbrush divide it. Dropped tris, really skinny tris.

I'll email you a file.
  Reply Reply More Options
Post Options
Reply as PM Reply as PM
Print Print
Mark as unread Mark as unread
Relationship Relationship
IP Logged

Previous
Next
 From:  Michael Gibson
1024.4 In reply to 1024.3 
Hi Crusoe, the dropped tris may be a bug, I'll check into that tomorrow.

Zbrush is adding a center point interior to the n-gon when it triangulates it - MoI does not add any interior points and just triangulates using points on the boundary. That seems to be the difference that you are talking about.

I think I can add that, I'll take a look. It doesn't completely guarantee removing skinny triangles, but it does stop them from traversing the entire n-gon and I think it is an improvement.

However, in your example you have diced everything into small bits using "Divide larger than" so that each n-gon is convex, this type of situation does make it easier to add just a center point for the triangulation. It is not easy though for larger more complex n-gons that are non-convex. I would be interested to see what ZBrush does with those types of n-gons, like if you leave "Divide larger than" off.

- Michael
  Reply Reply More Options
Post Options
Reply as PM Reply as PM
Print Print
Mark as unread Mark as unread
Relationship Relationship
IP Logged

Previous
 From:  Michael Gibson
1024.5 In reply to 1024.4 
The new patch available here: http://moi3d.com/forum/index.php?webtag=MOI&msg=1013.24

now performs this type of "connect to centroid point" triangulation when possible, it does seem to generate triangles that are a better proportioned.

Also I fixed a couple of bugs related to triangle optimization, that also should make a difference as well.

- Michael
  Reply Reply More Options
Post Options
Reply as PM Reply as PM
Print Print
Mark as unread Mark as unread
Relationship Relationship
IP Logged
 

Reply to All Reply to All