Another thing.

Did you notice that you can directly rendermap to another UV space?

E.g. shader using the huge projection and only a snippet of the UV space
rendermapped directly to a new 0-1 UV space.

This should even work for cases where you moved a tile segment to another
world grid location, as long as the UV projection stayed intact/where frozen 
already.

I like to create a square helper grid, create a new planar projection onto that,
then position the grid to my liking and just connect any other mesh to that 
grid´s
projection node to get identical UV spacings and scales for arbitrary objects.

With 15 pieces, this still seems doable and manageable.

In terms of using the full 0-1 space and pesky seams, using full 0-1 range
is nice because you could use Photoshop to paint out any sampling seams
on adjacent tiles easily but I´m not sure if filtering wouldn´t bite
you anyway? Mudbox for example keeps a percentage of the UV space
reserved for "edge bleding/edge extensions" on it´s default plane
and Ptex seems to do so as well? Only Fiddling and testing will give an 
answer...

Another reason for seams might be the mR meshsplittfactor settings, other´s here
reported nasty random seams depending on the mental ray meshsplitting settings 
picked.
That´s if you use mental ray at all...

Cheers,

tim



On 18.05.2012 13:54, Christian Gotzinger wrote:
Hi Tim,

Thank you for answering.
Unfortunately, I don't have access to ZBrush, but the idea of extending the 
color sounds good. I will try doing that in Photoshop.
I'll also look into premultiplication issues, but I don't think that's it.

Regarding the texture size, we are working with 15 by 15 tiles. So each tile receives 
almost 1,200x1,200 pixels. I say "almost" because the tiles don't coincide with 
the edges of
the huge texture. We had to adapt to a world grid, so the top left corner of 
the top left segment is not the top left corner of the huge texture.


On Fri, May 18, 2012 at 12:54 PM, Tim Leydecker <bauero...@gmx.de 
<mailto:bauero...@gmx.de>> wrote:

    Hey Christian,

    if you have access to ZBrush, you could load one segment and the
    Photoshop-cropped texture and try the "Fix seams" command.

    It惻l extend the color around the UVshell borders by 4-16px.

    This may be enough to hide the seams.

    Another thing that comes to mind is the backround color of your
    rendermapped texture segment.

    Are you using the original 18Kx18K px file and just the alpha coverage
    file for the Photoshop action? This should prevent "pseudo-premultiplied"
    effects when cropping, as you don愒 have black pixels as opposed to using
    the rendermapped texture (which will be black outside the surface area)
    as the cropping source.

    Btw, if I divide 18000px by 200, I end up with 90px that愀 a very small
    tile and sounds like loads of interpolation to start with anyway?

    Cheers,

    tim


    On 18.05.2012 12:19, Christian Gotzinger wrote:

        Hi list,

        We are dealing with a gigantic piece of environment. The ground has a 
texture of about 18,000x18,000 pixels applied to it. We are now in the process 
of dividing the entire
        area up
        into 200 quadratic segments.
        Thankfully, Softimage has allowed me to select all 10,000 objects at 
once and dice them up into quadratic segments, which I have then detached. All 
is good on that front.

        However, we would now like the 18K ground texture to be divided up into 
smaller segments as well. Currently all the quadratic cutouts have the same 18K 
texture applied but
        only use
        a fraction of their UV space. I want them all to use UV space 0 to 1 
and have a matching texture.

        This is what I tried:
        I rendermapped the segments (again I'm glad to be able to rendermap 
everything at once) and ticked the Coverage in Alpha checkbox. I can then use a 
Photoshop action to
        select the
        files' alpha channel, crop down to that selection and save the file. 
Then I can apply the new texture to a segment, delete its old UVs and add a new 
0 to 1 projection. This
        works,
        however I end up with seams of about 1 pixel width at the segment 
edges. Probably because the alpha coverage is not detailed enough, and 
Photoshop then crops incorrectly. The
        cropped files also have slightly different image sizes. Or maybe the 
rendermapping is not accurate enough?

        Has anybody ever done something like this? I feel like I'm 90% there, 
but those seams are a bit of a problem.


Reply via email to