[Bf-committers] Subsurf + UVmap, an alternative solution

[Jean-Luc Peurière]

Le 29 oct. 05 à 00:08, Frédéric van der Essen a écrit :

> Anybody who uvmapped a subsurfed object with lscm knows that there  
> is huge distortions on the result, sometimes on a level that is  
> unacceptable...
> I did a lot of tests and i identified two "bugs" responsible of  
> theses distortions...
>
> The first is that LSCM and other uv projections gets the original  
> cage vertices positions instead of the modified ones. This give  
> huge deformations on the "beveled edges" or in any cases where the  
> subsurfed vertices are far away from the cage vertices.
>
> So a solution would be to add yet another button in the modifiers  
> elements to "affect LSCM" This will solve 50% of the distortions,  
> and will be very usefull not only for LSCM and subsurf, but for all  
> the modifiers and all the projections algorithms
>
> The second problem (specific to Catmull-Clark) is that there is no  
> function to map accurately the New vertices created by subsurf into  
> the UVspace, and so for Catmull-Clark a simple subdivision  
> projection is used wich gives deformations in triangles.
>
> I looked on the web and it seems like besides the secret solution  
> of pixar, other apps do tricks like showing the vertices subdivided  
> once in the uvwindow, or using splines and other strange tricks
>
> But a temporary solution could be to UVmap the new vertices created  
> by Subsurf with LSCM. Basically it would
> - Pin the cage vertices
> - Lscm everything else
>
> This gives perfect results if the UVmap is created entirely with  
> LSCM, but it gives  average to very bad results if the UVmap is  
> intentionally distorted so it isn't a solution for all cases and  
> should be left as an option in the subsurf modifiers.
>
> It would at least give excellent results in most common cases and i  
> think it is worth trying, because there is no official and exact  
> solution to wait for.

For the first problem, I already reported that some times ago, and i  
fully agree on that.

http://jlp.nerim.net/dev/unwrap-on-subd.jpg shows clearly the  
problem, left is subsurfed, right is cage only. it is obvious cage  
mapping is good,
but result on subsurf is way sub par.

I would even not add a new button, as the initial cage is never  
needed imho.
I also think that  the amount of distortion induced by the first case  
is much more than the second.

Notice that it means that when you change the subsurf level, you have  
to redo the mapping.

For the second problem, i think it may be not even needed as sub- 
faces are regularly spaced, and so, should not induce much more  
distortion
once first problem is solved.

-- 
lukep