Re: [Algorithms] Lightmapping method

[Paul at H. <pa...@ru...>]

Thanks, I'll bear this in mind.  You guessed right about shadows. I've 
actually started doing that, but I've had to move on to redoing my octree 
stuff as it was a port from an old engine and there's some bugs in that too! 
Not a good weekend so far :(

Regards,
Paul Johnson.

Managing Director,
Rubicon Mobile, Ltd.
www.rubiconmobile.com

----- Original Message ----- 
From: "Thatcher Ulrich" <tu...@tu...>
To: "Game Development Algorithms" <gda...@li...>
Sent: Sunday, February 25, 2007 5:13 PM
Subject: Re: [Algorithms] Lightmapping method


> On 2/24/07, Paul at Home <pa...@ru...> wrote:
>> I have the basic idea about rasterising the triangle. I was trying to set 
>> a
>> texel box if the line passed anywhere through it as opposed to just the
>> right side of centre. I think I must've had some bugs or alignment issues
>> going on though as even after outline filtering it I still saw the
>> occasional gap or crack.
>>
>> I've switched over to my plan B after someone mentioned barycentric
>> coordinates to me and it all dropped into place. I now run over the 
>> entire
>> rectangle, use bary to get the right point/normal on the triangle and 
>> bobs
>> your uncle. I appreciate that with care a triangle plotter could/shoud be
>> workable, but this new (to me) method is a lot more robust. I get no 
>> seams,
>> no gaps, no anything and the new code took 10 minutes to slot in. Having
>> opened my post with a call for help, I'd now feel confident recommending
>> this method to anyone, if only for its ease of implementation.
>>
>> I didn't clamp those barycentrics btw, which is why it works so well, I
>> think. Sure the accuracy drops off as you start scanning outside the
>> triangle, but this fall-off is linear and gradual and the source normals
>> should all be fairly similar to begin with, so it's not noticeable in
>> practice. Not to me anway.
>>
>> A better still way would probably to switch to the "other" triangle as 
>> soon
>> as you pass out of the first one, but so far in my reasonably complicated
>> test case, I don't see anywhere where this refinement is needed - it all
>> looks great as it is :)
>
> 1. I'm guessing you're not computing shadows yet.  Smooth lighting can
> hide many fudges in a lightmapper, but shadow edges tend to expose
> them.
>
> 2. If and when you start mapping larger connected pieces of the mesh
> into one chart (as Jon mentioned), I think you will have problems with
> your technique, i.e. what do you do with interior edges?
>
> What worked for me was an antialiased renderer; i.e. compute exact
> texel coverage along boundaries.  I used an additive
> blending/precomputed alpha scheme for rendering, so the coverage went
> into the alpha channel.  After the chart was rendered, I bled unfilled
> texels using nearby filled texels (fattening).  And then, divide
> through by alpha to get the correct final color.  This makes interior
> edges and fattened texels behave pretty well.
>
> Another random trick (for the record): on the boundary of a chart,
> snap vertex mappings to texel centers.
>
> -T
>
>> Good point about using bigger polygons where possible. My current first 
>> pass
>> just puts one triangle into a rectangle and is very wasteful. I was 
>> thinking
>> about reflections into the 'other' corner to reuse space, but using 
>> larger
>> areas to begin with will help there too. I might need to look at my 
>> packing
>> routine a bit though so it's not dependent on outer bounding rectangles 
>> so
>> much as smaller divisions for "L" shaped polys.
>>
>> One other thing that overscanning back-projection may solve is the
>> differences between renderers. It's been a long time since I did any OGL,
>> but I remember at least at some point the texel offset rules were 
>> different
>> to D3D for example. This may of course apply to any other rendering
>> solutions too, so a triangle plotting solution would need to 
>> overcompensate
>> for this possibility too ?
>>
>> Regards,
>> Paul Johnson.
>>
>> Managing Director,
>> Rubicon Mobile, Ltd.
>> www.rubiconmobile.com
>>
>> ----- Original Message -----
>> From: "Jon Watte" <hp...@mi...>
>> To: "Game Development Algorithms" 
>> <gda...@li...>
>> Sent: Saturday, February 24, 2007 12:10 AM
>> Subject: Re: [Algorithms] Lightmapping method
>>
>>
>> >
>> >
>> > Paul Johnson wrote:
>> >> Hmmm, maybe I'm missing something here. When I have a triangle that's
>> >> basically a near-degenerate sliver at 45 degrees, all it's going to 
>> >> draw
>> >> is some dots. Fattening up just isn't working for me in those cases.
>> >> Maybe my UV are off generally, but I don't think so - I've spent time
>> >> looking at pixel centre issues already.
>> >>
>> >>
>> >
>> > If you rasterize the triangle as the union of the inside and its edge
>> > lines, and then fatten up, you'll get more than a few dots. Mapping
>> > backwards can work, too, as you say, but you have to clamp to the
>> > nearest point within the triangle. Also make sure to leave enough 
>> > gutter
>> > for MIP mapping. Because of gutters (and other filtering and efficiency
>> > reasons), good light mappers will attempt to create as large connected
>> > charts as possible for the light maps, btw.
>> >
>> > Cheers,
>> >
>> >          / h+
>> >
>> >
>> > -------------------------------------------------------------------------
>> > Take Surveys. Earn Cash. Influence the Future of IT
>> > Join SourceForge.net's Techsay panel and you'll get the chance to share
>> > your
>> > opinions on IT & business topics through brief surveys-and earn cash
>> > http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
>> > _______________________________________________
>> > GDAlgorithms-list mailing list
>> > GDA...@li...
>> > https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
>> > Archives:
>> > http://sourceforge.net/mailarchive/forum.php?forum_id=6188
>> >
>> >
>> >
>> >
>>
>>
>> -------------------------------------------------------------------------
>> Take Surveys. Earn Cash. Influence the Future of IT
>> Join SourceForge.net's Techsay panel and you'll get the chance to share 
>> your
>> opinions on IT & business topics through brief surveys-and earn cash
>> http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
>> _______________________________________________
>> GDAlgorithms-list mailing list
>> GDA...@li...
>> https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
>> Archives:
>> http://sourceforge.net/mailarchive/forum.php?forum_id=6188
>>
>>
>
> -------------------------------------------------------------------------
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get the chance to share 
> your
> opinions on IT & business topics through brief surveys-and earn cash
> http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
> _______________________________________________
> GDAlgorithms-list mailing list
> GDA...@li...
> https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
> Archives:
> http://sourceforge.net/mailarchive/forum.php?forum_id=6188
>
>
>
>