Camera-Space Cubemap Faked AO

[LEGION]

Ah, now you have your participants! I was able to follow a little bit, but it would not hurt if you could add more screenshots, but I guess it´s still all theory?

And with GI you mean Global illumination?

[STiFU]

Yes Legion.

 

By the way, I just found a nice presentation NVidia held on siggraph about "Horizon-Based Ambient Occlusion" (short: HBAO), where surface normals are computed from the sampled depths through atan. This is probably one of the newer approaches Rebb has been talking about and it is Sikkpin's newest project. We'll soon see how that compares to the default SSAO in Doom 3. I wonder whether he also used all the optimizations proposed in the respective paper in his SSAO implementation.

[LEGION]

Erm, Doom3 had SSAO? Not only AO? At least Doom3 does not appear on this list here.

 

I just installed Amnesia: The Dark Descent and noticed the SSAO-settings, meaning the Sample-setting. Though I have a descent system, the menu stutters, well the mouse movement does, when I chose the maximum with 128. Now I set it to 32 (resolution set to high) and I hope I won´t have stutter in the game, the menu works fine. I won´t start the game until I finished Penumbra Series.

 

So, how much samples does Doom3/TDM use? Is it related to this: r_multiSamples "4" ?

[nbohr1more]

SSAO requires Sikkpin's HDR mod:

 

 

http://www.doom3world.org/phpbb2/viewtopic.php?t=22822

 

(D3W topic)

[LEGION]

Oh builder, it´s pretty heavy! Damn, on that shots I saw monsters that I didn´t saw in Doom3 yet. Thanks, Nils.

 

Oh no!!! I can´t download the rar-file.

Edited September 22, 2010 by LEGION

[nbohr1more]

Towards the end of the thread I believe there are links... But it may also be at filefront (haven't had a chance to check yet)...

[LEGION]

No, but the first post was edited recently...

 

On the filefront I only found TRSGM's HDR Interaction Redux: The Bumpening Edition.

[nbohr1more]

The 1st post D3W rar link works for me... Dont know if it has the newer SSAO though...

 

Edit:

 

This one?

 

http://sikkpin.fileave.com/sikkmod_test5-D3.rar

Edited September 22, 2010 by nbohr1more

[LEGION]

What? I still get an server-error. No matter how I try. (save target or click)

[nbohr1more]

Looks like someone at guru3d also posted a rapidshare:

 

http://forums.guru3d.com/showthread.php?p=3712139

[LEGION]

This one?

 

http://sikkpin.filea...od_test5-D3.rar

 

Yes, that gives me: "Fehler: Verbindung fehlgeschlagen", that´s german for: jerks like you are not allowed, go home."

 

Thanks for the RapidShare, now there I have the same trouble as with Baddgogs mission, I tried first with my download-plugin and it gives me the html-file. ARGH, now I can wait 15 minutes to try again and I hope that I won´t forget to just click 'Save target'...

[STiFU]

Haha 128 Samples for AO calculation in Amnesia is overkill. I used 8 samples at high AO-resolution and the results as well as performance were fine.

 

The SSAO dependent settings in sikkmod start with r_SSAO. I can't remember the exact settings though and I am also not at my home computer to look them up. The multisample setting is for Anti Aliasing I believe.

[LEGION]

Hehe, I don´t have a clue what that samples are for anyway...

 

Yeah, I think you´re right about that it means AA, it´s the same value I chose.

 

And the download was finally a success. Hmm, I think it´ll turn out in a mess when I try to merge this into the DentonMod I already messed around with a lot. Ah, it can´t work because of the altered dll...

 

For some reason I can´t get AA and my custom res enabled, what I need is r_mode=-1, but constantly I fall back to 640x480, oh well.

 

Also it looks really blurry, but I need to play around with the values.

 

But now I want to go to the bathroom and then I go into the forest from NHAT, that looks as stunning as SSAO, doesn´t it?

[nbohr1more]

Yeah, NHAT 3/3 is very pretty...

[STiFU]

Hehe, I don´t have a clue what that samples are for anyway...

Normally the AO value would be computed with all the depthvalues surrounding the current pixel inside a certain range. This would be way too expensive computationally of course, so to reduce the calculations a random kernel is used to pick a reduced number of samplepixels arount the current pixel. You should generally choose a value of 16 Samples per calculation, as this is said to be a good value in numerous papers.

[LEGION]

Ah thank you, then 16.

[nbohr1more]

Now that I've gotten this silly hack idea out of my system, I have tried to fathom some of Rebb's work in the ambient fix.

 

I see that the current implementation essentially looks for an RGB 01 value from any (?) projection image's alpha to essentally say "Don't paint ambient light here" (?)

 

I see that an experimental build is sampling two "Light Normalization" cubemaps. Which means that each face of the cubemap is the average inbound light to the player position?

 

Could this system be made to dynamically accept more detailed (pre-rendered) "scene" cubemaps that would be weighted against the normalization cubemap depending on the player position to the scene?

 

All the mapper would then have to do is collect an envshot (with lights added to simulate bounces) and plug the cubemaps and positions into "the system".

 

(Maybe you guys are already cooking-up something akin to this...)

 

(Probably missing some clue here anyway )

[nbohr1more]

Clue found:

 

The Normalization Cubemap is a mathematical structure used to help bypass hardware limits for normal vector angle calculations.

 

(Therefore has no direct bearing on precomputed cubemap based "Image-Space Lighting" solution approaches... ...)

 

More studying...