Pages: 1 2 [3]   Go Down
  Print  
Author Topic: SSAO TechDemo - Update 25.08.2010  (Read 6796 times)
May 06, 2010, 02:18:10 pm
I think I'll leave the implementation to those with an actual degree in programming or math, thank you very much Wink

Anyway, it's good to know that this wasn't overlooked - the results look great so I guess the future of high quality AO in realtime is very bright indeed Smiley
May 06, 2010, 02:59:51 pm
viktor, did you make a demo from your solution? even if it is slow i would love to see the results.
May 06, 2010, 04:24:36 pm
Nope, I just look at their source and made an simulation with aproximatetly the same number of instuctions and samplings from texture. As to issues with normals I hit that when tryed to include normals into my ssao. Building a separate buffer with a "flat normals" is possible with MRT, but basically you just waste a memory as you won't use these normals anywere else than for ssao.

Their original solutions dosn't scale sampling with a distance and that's another reason why it's slow. They take samples with pretty big radius around current fragment, this treshes caching and as closer to the surface you set a camera the worse it becomes, as it starts to take samples over the whole screen.

AO is very sabtle effect and it should be as fast as possible, take a look at Unreal's SSAO, it looks much worse than what we have, but used for AAA games.

Upgrade to PSSM is available for SSAO customers. Check http://www.3dvrm.com/shadows_solution/ for details.
May 06, 2010, 04:28:54 pm
I see the same mistake over and over again - when somebody starts to implement ssao they look at it's pure result - grayscale image. They trying to make it as smooth and detailed as possible, forgeting that in real scenario, diffuse textures and caluclated lighting will mostly vanish ssao term and nobody will really see how that implementation is better than another one, until you start to point finger at some specific points.

Upgrade to PSSM is available for SSAO customers. Check http://www.3dvrm.com/shadows_solution/ for details.
May 14, 2010, 08:11:45 pm
Update: SSAO ver.0.9
Model Stanford Dragon - 869K poly model.


* 1.jpg (137.12 KB, 1394x976 - viewed 202 times.)

* 2.jpg (94.92 KB, 1398x995 - viewed 212 times.)

* 3.jpg (67.93 KB, 1393x989 - viewed 200 times.)

* 4.jpg (117.29 KB, 1378x986 - viewed 207 times.)
May 14, 2010, 08:27:50 pm
I see the same mistake over and over again - when somebody starts to implement ssao they look at it's pure result - grayscale image. They trying to make it as smooth and detailed as possible, forgeting that in real scenario, diffuse textures and caluclated lighting will mostly vanish ssao term and nobody will really see how that implementation is better than another one, until you start to point finger at some specific points.
That might go well for a game where the designers can just design the diffuse textures, but in architectural visualizations there might be plenty of solid colors. (Window frames to start with and they are a perfect candidate for SSAO.) By just optimizing your sampling strategy you can improve the smoothness without increase in computation time, so I think that's always worth the trouble. Indeed you'll have to keep the sampling range low, but that's where light baking and SSAO complement each other. I only wish that I could render my depth image larger than the viewport to prevent any artifacts at the borders of the screen. (Again, at minimum expense.)
May 15, 2010, 06:07:21 am
Demo: SSAO ver.0.9 - Model Stanford Dragon (~24 MB)
Demo2: SSAO ver.0.9 - Model Sponza Atrium (~4.5MB)
Test. If can write how many at you FPS. Smiley

* SSAO2.exe (4553.96 KB - downloaded 403 times.)
August 24, 2010, 07:21:08 pm
Update: SSAO ver.1.1 Final  Cheesy

* ssao.exe (3977.08 KB - downloaded 226 times.)
August 27, 2010, 11:18:28 am
Really cool Result. I really like the result. I test it with my own model. Is it also support alpha map (Plants)?

ali-rahimi.net
August 27, 2010, 12:07:38 pm
yes this version support alpha map. Cheesy
August 27, 2010, 02:21:29 pm
I do a test to compare Viktor VS Anuar SSAO  Grin. Top image is Anuar's occ and below is for Victor. different is more when we get closer.


* occ-1.jpg (297.83 KB, 1920x2260 - viewed 171 times.)

* occ-2.jpg (368.96 KB, 1920x2260 - viewed 175 times.)

ali-rahimi.net
August 27, 2010, 02:35:25 pm
so, which result you like more and what about performance?
August 27, 2010, 02:38:01 pm
Anuar's look better. About the performance when i zoom in with Anuar's occ fps drop. Dont know why.

ali-rahimi.net
Pages: 1 2 [3]   Go Down
  Print  
 
Jump to: