Bow down before your new master!

Share with us how are you using the powerrrr of the force

Moderator: Alain

Postby agi_shi » Sun Apr 20, 2008 12:06 pm

:shock:

Wow. That definitely looks very nice. I really have only 1 question: what shadow mapping technique are you using for the soft shadows? Wild guess, but layered variance shadow maps?
agi_shi
 
Posts: 263
Joined: Fri Aug 17, 2007 6:54 pm

Postby Leadwerks » Sun Apr 20, 2008 3:46 pm

I don't know why people even talk about variance shadow maps. You can just take four shadow samples and the results are much better then VSM.
User avatar
Leadwerks
 
Posts: 569
Joined: Fri Oct 27, 2006 2:54 pm

Postby agi_shi » Sun Apr 20, 2008 4:38 pm

Leadwerks wrote:I don't know why people even talk about variance shadow maps. You can just take four shadow samples and the results are much better then VSM.

Layered variance shadow maps are as close you'll get to complete soft shadows as currently possible. 4 shadow samples, what do you mean? Brute-force PCF? There is no way PCF can look anything like your screenshots, especially not with 4 samples.
agi_shi
 
Posts: 263
Joined: Fri Aug 17, 2007 6:54 pm

Postby Leadwerks » Sun Apr 20, 2008 4:53 pm

4 samples is hardly brute force on modern hardware.

Yes, I just use four samples for each fragment. I don't know why other programmers make such a fuss about how hard it it.
User avatar
Leadwerks
 
Posts: 569
Joined: Fri Oct 27, 2006 2:54 pm

Postby agi_shi » Sun Apr 20, 2008 5:13 pm

Leadwerks wrote:4 samples is hardly brute force on modern hardware.

Yes, I just use four samples for each fragment. I don't know why other programmers make such a fuss about how hard it it.

4 samples have only 4 levels of shadowing - 0% (unlit), 25%, 50%, 75%, and 100% (lit). Your screenshots show more than 4 levels of shadowing. It's not hard, it's just ugly. And your screenshots are much the opposite of ugly.
agi_shi
 
Posts: 263
Joined: Fri Aug 17, 2007 6:54 pm

Postby Leadwerks » Sun Apr 20, 2008 5:22 pm

NVidia cards have an automatic PCF filter, so that gets applied to each sample.

For ATI cards, you have to do some random grid rotation nonsense. I think I am going to toss that and do my own PCF filter based on texel position and weigh four pixels.
User avatar
Leadwerks
 
Posts: 569
Joined: Fri Oct 27, 2006 2:54 pm

Postby agi_shi » Sun Apr 20, 2008 9:08 pm

Leadwerks wrote:NVidia cards have an automatic PCF filter, so that gets applied to each sample.

Oh, well that's different ;). That makes it 4x4 which is is a 16-tap PCF kernel, which isn't too bad. Still, far from being "better than any VSM stuff" ;).
For ATI cards, you have to do some random grid rotation nonsense. I think I am going to toss that and do my own PCF filter based on texel position and weigh four pixels.

There's many ways to do PCF, do as you wish. Though, trust me, variance shadow maps (layered variance shadow maps these days) are a lot more than you seem to think of.
agi_shi
 
Posts: 263
Joined: Fri Aug 17, 2007 6:54 pm

Postby Julio Jerez » Sun Apr 20, 2008 11:25 pm

I think Agi shy is right.
I do not know the detail of the variance shadow map are, but I know of someone that implemented it for a XBox360 game after several attend at implementing 3 x 3, 4 x 4 filters, and the variant map was quite impressive and fast. Again I do not know any of the detail I just saw the result of the two techniques.
I think what Leawerk is doing is much better, buy I would not dismiss the technique so easy unless you had tried.
Julio Jerez
Moderator
Moderator
 
Posts: 12249
Joined: Sun Sep 14, 2003 2:18 pm
Location: Los Angeles

Postby Leadwerks » Sun Apr 20, 2008 11:29 pm

I thought VSM created light borders when the shadows of two objects overlap?
User avatar
Leadwerks
 
Posts: 569
Joined: Fri Oct 27, 2006 2:54 pm

Postby bakura10 » Mon Apr 21, 2008 4:48 am

There are so much techniques for shadow maps... I like it because the names are really long : Layered Variance Shadow Maps for instance (http://forum.beyond3d.com/showthread.php?t=47427).

Some techniques looks promising too. Have you get a look at Exponential Shadow Maps ?
"Hay othas optimizationes that make collison faster and also un crash bug fixed in el fracture code." Spanglish, Julio Jerez
bakura10
 
Posts: 30
Joined: Sun Sep 17, 2006 11:39 am
Location: France

Postby agi_shi » Mon Apr 21, 2008 9:44 am

Leadwerks wrote:I thought VSM created light borders when the shadows of two objects overlap?

That's call light bleeding. The "normal" VSM implementation suffered from quite a lot of light bleeding in complex scenes. But, like I mentioned, have you looked at layered VSM? That solves just about all of the light bleeding.

Look, I think your shadows are very pretty (or at least they look pretty in the screen shots ;)) - I'm just saying that PCF will hardly achieve what your screenshots look like, which is very nice.
agi_shi
 
Posts: 263
Joined: Fri Aug 17, 2007 6:54 pm

Postby Leadwerks » Mon Apr 21, 2008 1:22 pm

I'm just saying that PCF will hardly achieve what your screenshots look like, which is very nice.

I will look into LVSM for ATI cards, but I assure you I am only using 4 samples with the free PCF filter, and that the shots above are real-time (150-300 FPS now).
User avatar
Leadwerks
 
Posts: 569
Joined: Fri Oct 27, 2006 2:54 pm

Postby Vsk » Mon Apr 21, 2008 1:23 pm

I didn't know that phong can be acchiev in real time!.
I had phong at the same level (of cpu consuming) that radiosity (gathered) and raytracing.
The phong you are using is a varian from the normal phong to be used in real time?

And I agree that the tiles are very brights. Most modern games made everything very bright, I don't understand why :?. It looks less natural.

I will spect for the video :D.
Keep it up!.
Vsk
 
Posts: 34
Joined: Thu Oct 18, 2007 3:13 pm

Postby Vsk » Mon Apr 21, 2008 1:36 pm

Seen the pictures, I wonder, what about the physics?
Can a bullet pass through that collum that is quite smashed, int part broken?
(I know I bad, I know :P):
Vsk
 
Posts: 34
Joined: Thu Oct 18, 2007 3:13 pm

Postby bakura10 » Mon Apr 21, 2008 3:12 pm

Phong can run in real time, and per pixel, of course :p. It runs on my very old card (9800 Pro), and with my new card (HD3870), it runs so much better. Phong lighting is not that expensive now. Some papers uses a sort of instant radiosity for approximate indirect lighting : a lot of secondary light sources are created for simulating indirect light sources. On a 8800 GTX, I've read a paper that was able to runs more than 150 point lights with full Phong per pixel lighting at about 30 fps.
"Hay othas optimizationes that make collison faster and also un crash bug fixed in el fracture code." Spanglish, Julio Jerez
bakura10
 
Posts: 30
Joined: Sun Sep 17, 2006 11:39 am
Location: France

PreviousNext

Return to User Gallery

Who is online

Users browsing this forum: No registered users and 13 guests