Post Your Irrlicht Screenshots / Render Here.
Deferred CSM. (port from http://www.gamedev.net/community/forums ... _id=372739)
Last edited by wing64 on Fri Jul 16, 2010 3:41 am, edited 2 times in total.
Wow! When i read "CSM" i thought it was "Cascaded Shadow Maps".
Nice Parallax Mapping... or relief mapping... whatever it is called, it is great Though the triangle in the wall still looks funny... But i think that's not your code, but the tangent space not calculated properly.
never the less. Looks impressive
Nice Parallax Mapping... or relief mapping... whatever it is called, it is great Though the triangle in the wall still looks funny... But i think that's not your code, but the tangent space not calculated properly.
never the less. Looks impressive
"There is nothing truly useless, it always serves as a bad example". Arthur A. Schmitt
@stefbuet:
Thanks.
@Mel:
You right, Mel. Sometime when i load mesh all extension supported by irrlicht except .3ds, Engine shown some face wrong. But when i run binary code from http://www.gamedev.net/community/forums ... _id=372739 (He use irrlicht 1.4 mod) it work perfectly. I very confuse about that.
Thanks.
@Mel:
You right, Mel. Sometime when i load mesh all extension supported by irrlicht except .3ds, Engine shown some face wrong. But when i run binary code from http://www.gamedev.net/community/forums ... _id=372739 (He use irrlicht 1.4 mod) it work perfectly. I very confuse about that.
Since 3ds splits all the vertices I'm guessing it's meshes with vertices(normals) shared across triangles that cause problems. Try the effect on a sphere with smooth normals then a 3rd sphere and see if the effect still looks fine without normal smoothing. (could also be the tangent basis is generated differently in the guys custom irr)
Mel: those effects look really nice, is that a sphere mapped terrain? looks abstract
Mel: those effects look really nice, is that a sphere mapped terrain? looks abstract
"Irrlicht is obese"
If you want modern rendering techniques learn how to make them or go to the engine next door =p
If you want modern rendering techniques learn how to make them or go to the engine next door =p
Deferred lighting with normal map + alpha support + light frustum culling + bloom. I still have to do real motion blur, and antialiasing (edge detection + blur edge like stalker does [see gpu gems] + shadows)
I still don't know what shadow kind I'll choose. Dual paraboloids shadow mapping looks pretty cool Mel, but if you have to tesseleate the scene, it's not that cool because I wanted do have per pixel lighting to handle huge polygons which fail with per vertex.
This is the bad part I guess
There are 50 lights, 32000 polygons, 512x512.
I run on a Geforce 260GTX EXO and a Phenom II X3 710.
The part which cost the most is multipass.
I mean with one light I got 550FPS, with 50 lights only 115.
I think it's because of my RenderTargets. I use 5 render targets A16R16G16B16 because 8bit per canal was not enought precision. However I use multiple render targets.
Here is my Render Target usage :
(note there is a mistake, there is not Texture Alpha component, it's an empty case, so I got 16bit per pixel wasted instead of 8 )
I'm not really satisfied with this performence because few weeks ago I ran a demo from a guy with deferred normal mapping and 1024 lights @ 200 FPS !! insane, I'm really jealous and even more because I can't find again the URL of the demo & sources. I guess this demo was a quad textured only...
But hey, I've just tested to render the scene using basic smgr->draw method and I got only 250FPS when looking at the room, about 1000 when the room is removed thanks to the frustum culling. That's fun my shader run faster when using 1 light or little more.
Tell me what you think
There are 50 lights, 32000 polygons, 512x512.
I run on a Geforce 260GTX EXO and a Phenom II X3 710.
The part which cost the most is multipass.
I mean with one light I got 550FPS, with 50 lights only 115.
I think it's because of my RenderTargets. I use 5 render targets A16R16G16B16 because 8bit per canal was not enought precision. However I use multiple render targets.
Here is my Render Target usage :
(note there is a mistake, there is not Texture Alpha component, it's an empty case, so I got 16bit per pixel wasted instead of 8 )
I'm not really satisfied with this performence because few weeks ago I ran a demo from a guy with deferred normal mapping and 1024 lights @ 200 FPS !! insane, I'm really jealous and even more because I can't find again the URL of the demo & sources. I guess this demo was a quad textured only...
But hey, I've just tested to render the scene using basic smgr->draw method and I got only 250FPS when looking at the room, about 1000 when the room is removed thanks to the frustum culling. That's fun my shader run faster when using 1 light or little more.
Tell me what you think
Hi,
Thanks for good information about your work. And below image shown deferred in my framework. I add some technique e.g. VSM, CSM and Blinn-Phong, Shader run on i7-930, Radeon HD4600.
Framwork use RTT size 1024x1024.
- Diffuse A8R8G8B8
- Normal A8R8G8B8
- Depth R32F
- Shadow Depth G32R32F
- Point light rendering with Quad. (i'm not sure speed faster between both quad or sphere )
Thanks again for good information.
Result ->
1 light
50 light
1 light
50 light
Thanks for good information about your work. And below image shown deferred in my framework. I add some technique e.g. VSM, CSM and Blinn-Phong, Shader run on i7-930, Radeon HD4600.
Framwork use RTT size 1024x1024.
- Diffuse A8R8G8B8
- Normal A8R8G8B8
- Depth R32F
- Shadow Depth G32R32F
- Point light rendering with Quad. (i'm not sure speed faster between both quad or sphere )
Thanks again for good information.
Result ->
1 light
50 light
1 light
50 light
wing64 and stefbuet: LOOKS AWESOME!! Next you'll need some AA to make it look even better
My company: http://www.kloena.com
My blog: http://www.zhieng.com
My co-working space: http://www.deskspace.info
My blog: http://www.zhieng.com
My co-working space: http://www.deskspace.info
There's a problem with doing anti-aliasing when rendering to texture. Hybrid showed me a workaround here for my postprocessing library. Basically you render to the frame buffer and then use DX or OGL code to copy it to an RTT.
Hey wing64 nice screen shots
But I'm wondering, don't you use multiple render targets?
Because when using multiple render targets they have to be the same format (bit/pixel). my precision problem was with the depth map. When I started I took a simple R32 render target and A8R8G8B8 for other maps, but then when using multiple render targets feature I had to choose an R16G16B16A16 for everything because of the depth... Maybe I could do this in 2 passes, 1 for depth R32 and R8G8B8A8 for all others in other pass idk if it will go faster.
I'm currently trying to do some antialiasing. An edge detection shader working with depth and normal difference beween pixels. Then I will apply a blur on edge detected pixels.
Again nice work!
Your 3D model reminds me of that : http://www.gamedev.net/reference/articl ... le2193.asp
But I'm wondering, don't you use multiple render targets?
Because when using multiple render targets they have to be the same format (bit/pixel). my precision problem was with the depth map. When I started I took a simple R32 render target and A8R8G8B8 for other maps, but then when using multiple render targets feature I had to choose an R16G16B16A16 for everything because of the depth... Maybe I could do this in 2 passes, 1 for depth R32 and R8G8B8A8 for all others in other pass idk if it will go faster.
I'm currently trying to do some antialiasing. An edge detection shader working with depth and normal difference beween pixels. Then I will apply a blur on edge detected pixels.
Again nice work!
Your 3D model reminds me of that : http://www.gamedev.net/reference/articl ... le2193.asp
My game project:
My company: http://www.kloena.com
My blog: http://www.zhieng.com
My co-working space: http://www.deskspace.info
My blog: http://www.zhieng.com
My co-working space: http://www.deskspace.info