Sharp Edges in One Smoothing Group [UDK/3DS Max]

Maecaune

Hi there,
I am working on some kind of temple and I ran into a problem. In order to save texture-space I am using tilable textures, but I want a clearly visible edge that is not sharp as a knife.

The only solution I have right now is to add additional edges next to the corner using one Smoothing Group. Visually the result is great, but now I have alot of more triangles.

I made a simple test to show you what I mean. I wonder if there is another solution or are additional edges the way to go?

ZacD

Face weighted normals might help, but I don't know how well they with with UDK

http://www.polycount.com/forum/showthread.php?t=85809

m4dcow

On the 1st wireframe, you can get rid of the loops on the left and right, and use face weighted normal/edit the vertex normal to achieve the look you are going for.

With tileables, there isn't really any other option to do what you are looking to do without adding geometry.

I use maya, and a tool called migNormals tools to edit my vertex normals.

Here is a quick example I just did, weighs in at 92 tris vs 108, with yours, not a big difference, but same fidelity at a reduction of tris.

Maecaune

Thank you for you great answers, but it seems I can't achieve it that fast wird 3DS Max. Given the temporary time-limit I have for the project right now, I think I drop the idea for now using weights. It looks like Maya is using it as a standard, but 3DS Max don't? This weight-painting in Max looks a lot more complicated, than hoped for it to be.

But on my journey I luckily found the solution to my initial problem. Editing normals solves trouble of using triple-edges.

Bek

You might find this thread interesting: http://tech-artists.org/wiki/Beveling
(And possibly also http://www.polycount.com/forum/showthread.php?t=106108)

So it's the vert count that matters rather than tri count.

My understanding is a vert can only have one normal. So when there's a hard edge, say at the edge of the cube, the 1 vert becomes 3 as there's a normal for the x,y,z of the cube's faces. Additionally, a vertex can only have one uv coord, so if there's a uv seam (1 vert having 2 uv co ords) that's a split too - unless it's already split, the splits don't stack that way.

So for your first example, adding edge loops where there isn't a uv seam doesn't end up with more verts than you'd first think. For example: both these shapes are 24 verts in editor:



The difference is quite small on a simple asset like this; denser models will obviously benefit greater from a more conscious use of bevels. In this case the simple bevel has less splits (because of the averaged normal, instead of split normals to produce a hard edge) so only has a minor increase in total vert count. Hopefully this explains why tri count can be deceptive.

Also there are some good reasons for/against weighted normals if you're interested: http://www.polycount.com/forum/showthread.php?t=85809

Maecaune

That is an really interesting approach to the vertex- and tri-count. Thinking about it, a professional artist told me once, that in reality the engine splits the meash between smoothing groups. So it makes sence to me. I will take this into consideration.

Thanks .