High Poly Count aliasing ? back

(L) [2007/09/24] [Shadow007] [High Poly Count aliasing ?] Wayback!

Phantom, I re-read your paper this Week-End, and noticed one particular paragraph :
(L) [2007/09/24] [Phantom] [High Poly Count aliasing ?] Wayback!

The detailed models my students used contained many extremely fine details. The aliasing on such details is very noticeable. By replacing this detail with normal maps, the details are at least filtered. This improved matters. If you approximate smooth surfaces with lots of triangles, this is not a problem; it is only a problem with extreme detail.
_________________
--------------------------------------------------------------

Whatever
(L) [2007/09/24] [Shadow007] [High Poly Count aliasing ?] Wayback!

I thought that RT would mean techniques like parallax mapping would be replaced by the large poly count "original"...  [SMILEY Sad]
(L) [2007/09/24] [ingenious] [High Poly Count aliasing ?] Wayback!

Yes... geometric aliasing becomes an issue when you can afford detailed geometry. When each ray hits a different triangle, the image doesn't look that pretty anymore. You either need to do super-sampling (which scales nearly linearly unfortunately) or to prefilter the geometry. That is why the low-poly model helped. We should really think about level of detail...
(L) [2007/09/24] [Wussie] [High Poly Count aliasing ?] Wayback!

[IMG #1 ]

[IMG #2 ]


visuals to support phantom's point! The aliasing clearly is most noticeable at the edge of geometry (duh) but I believe most of it is visible near the edges of the floor tiles and the green 'air-ducts' cause of a cluttering of geometry over there. Ofcourse with clever thinking this could've been replaced by a normal or bump map... But that aside [SMILEY Wink]
_________________
[LINK http://www.lost.eu/66d97 Whatever you do, don't click me]
[IMG #1]:Not scraped: https://web.archive.org/web/20071026162646im_/http://dump.harbl.nl/aliasing.png
[IMG #2]:Not scraped: https://web.archive.org/web/20071026162646im_/http://dump.harbl.nl/aliasing2.png
(L) [2007/09/24] [goodbyte] [High Poly Count aliasing ?] Wayback!

I can even stretch it as far as saying that geometric aliasing is one of the biggest problems to solve right now, yes there are numerous attempts (geometry caches, brick maps, point clouds, etc.) but each has it's drawbacks. Once we have good geometry filtering (and a traversal structure to match it) we should be able to handle incoherent rays (f.i. glossy reflections and ambient occlusion) in much better ways, both quality and speed wise.
(L) [2007/09/25] [Phantom] [High Poly Count aliasing ?] Wayback!

Indeed, ingenious. Look at the green ceiling objects, these used to be twice as detailed, resulting in problems even at high resolutions. Normally, students don't run the game at 1024x768. [SMILEY Smile]
_________________
--------------------------------------------------------------

Whatever
(L) [2007/09/25] [ingenious] [High Poly Count aliasing ?] Wayback!

The scene looks pretty nice though [SMILEY Smile] What's this blurrish effect?
(L) [2007/09/25] [Wussie] [High Poly Count aliasing ?] Wayback!

I believe it's caused by Phantom's HDR implementation in combination with the "Emissive" material hack we suggested during the production process. I'll probably be experimenting with some anti-aliasing techniques during the next weeks, if something decent pops up i'll let you know.
_________________
[LINK http://www.lost.eu/66d97 Whatever you do, don't click me]

back