Triangle Strip Preserving LOD (T-Strip LOD)

Triangle Strip Preserving LOD (T-Strip LOD)

Post by Ernest Szok » Fri, 10 May 2002 03:35:54



Hey Guys

I have a paper and demo on a new alternative to ROAM and other CLOD
algorithms for terrain LOD. I call it T-Strip LOD and it basicly subdivides
complete lines instead of triangles, esentially giving you the same
structure of a quad tree, except that blocks form a contigious line across
one axis which give you the contigious trinagle strip. It's relativly simple
and takes advantage of the high processing power of todays hardware
graphics, but with less CPU usage than conventional LOD algorithms. Perhaps
not as effective as VIPM, it may still prove to be relavent in certain
environments where there is deforming terrain, extensive terrain collision,
and mapping objects to the visble terrain mesh is required. It's more of a
cross between brute force triangle strip rendering  and CLOD.

Anyhow, the paper and demo are up at
http://chat.carleton.ca/~eszoka/tstriplod/tstrip.htm, note the demo takes 30
seconds or so to generate the terrain texture, read the readme.txt for the
keys to operate the various features and controls.

Any thoughs or comments would be appreciated.

Ernest Szoka
http://chat.carleton.ca/~eszoka/

 
 
 

Triangle Strip Preserving LOD (T-Strip LOD)

Post by Cruz Vincen » Sat, 11 May 2002 18:51:15


Hi,

  The demo is quiet nice. Last year i tried to implement full-ROAM technique
but making it hardware friendly was painfull (i didn't manage to :) ). It seems
to be a nice alternative to other HW friendly LOD techniques (geomipmap).

--
Cruz Vincent

 
 
 

Triangle Strip Preserving LOD (T-Strip LOD)

Post by bob » Tue, 14 May 2002 04:26:09


Interesting demo.

Unfortunatly still get the morphing effect as the lines pop !

Be nice if there was source code .

brian.


Quote:> Hi,

>   The demo is quiet nice. Last year i tried to implement full-ROAM
technique
> but making it hardware friendly was painfull (i didn't manage to :) ). It
seems
> to be a nice alternative to other HW friendly LOD techniques (geomipmap).

> --
> Cruz Vincent

 
 
 

Triangle Strip Preserving LOD (T-Strip LOD)

Post by Seppo Citymarke » Tue, 14 May 2002 17:06:33


Quote:> Any thoughs or comments would be appreciated.

Is it supposed to run 4 fps on Pentium4 1.7Ghz + GeForce3 ? For real-time
rendering this is a bit too slow, unless there is something wrong with my
graphics card drivers / configuration. Is this normal? At 4 fps my
configuration can render 7M triangles per frame (with brute-force through
DirectX 8.1), so I would expect a bit higher framerate for custom algorithm
optimized for performance (?).

--
SCM

 
 
 

Triangle Strip Preserving LOD (T-Strip LOD)

Post by Seppo Citymarke » Tue, 14 May 2002 22:53:29


Quote:> It runs at about 85 FPS, doing 270K triangles per second on my machine
> (Athlon 1800+ 1.5GHz with Geforce 3). In my opinion it looks very good,
> better than what many commercial games uses.

Then my configuration is broken somewhere (regarding OpenGL atleast).

Quote:> The 28M triangles/sec figure is just marketing crap.

They advertise MUCH higher figures in the "marketing crap".

Quote:> Maybe you can achieve this drawing 1 static mesh with
> 28M triangles onto a 1x1 rendertarget with no textures.

I seem to get this performance constantly in "production quality" scenes.
Then again, our models re-use same texture very frequently, we use the "blow
mesh open, planar map, morph back to original morphtarget" -mapping so that
most of the times we get away with one texture per object (assuming that
have enough texels). This reduces number of state changes significantly.
When doing translucent primitives, we determine if the scene can be layered
(say, water, no point bringing translucent water primitives through the
translucency/sorting pipeline).

Quote:> Drawing a terrain with CLOD and materials is something completely

different.

I never said it was. I use quadtree myself with pre-tesselated geometry
mipmaps "geomips" and geomorph between the levels where there is transition,
most of the primitives going through this pipeline are static. The goal is
only to have high range of visibility, equivalent of 16K x 16K grids are
piece of cake for arrangement like this as far as fps is concerned. Storage
is another matter, dataset with this size don't fit into memory at full
quality so we have to page. Textures are parametrically defined so higher
quality "unique" textures are quick to calculate into the cache when needed.

28M triangles/sec isn't marketing bullshit, it shoud be the minimum goal to
aim since that's what (this) hardware seemingly can easily do. Just my
$0.02, comments were asked and I gave some based on my own experience in the
matter, it wasn't flame if you wondering. I want to stress that I just found
it bizarre that the framerate was this low, if the work seemed to be
concetrated on *performance* optimization - why else would so much care put
into output of triangle strips). That's all. Now I know it was my system.
Thank you.

--
SCM

 
 
 

Triangle Strip Preserving LOD (T-Strip LOD)

Post by Seppo Citymarke » Tue, 14 May 2002 23:27:40


Quote:> Then my configuration is broken somewhere (regarding OpenGL atleast).

... which was! I was using 23.11 detonators because the nv refresh rate fix
works with them (all fullscreen modes, where application don't allow to
override, don't synchronize). Anyway, nVidia could address this oneday in
their drivers - meanwhile, I guess have to either:

<nvidia nitpick>

1. Upgrade to XP, since XP SP1 coming out in fall (LOL) will address the
issue (Microsoft atleast does something - too bad such fix isn't forthcoming
to 2000)

2. Upgrade to LCD flatpanel display (currently using 22" 1200NF which is
very nice display - except - warning - with nVidia hardware ;-)

</nvidia nitpick>

-- tomorrow Matrox will officially announce their Parhelion -- check it out.
;-)

Anyway, with 29.20 detonators the framerate is 85 (screen refresh is 85hz,
so I take it's maxed out). This gives much better impression of the routine,
so I thereby withdraw my criticism. What hits me with odd-bat is that my
own, and other OpenGL applications didn't suffer from this "framefrate
lowness" syndrome with 23.11's, only this demo application. Was some GL
extension used which wasn't available in older drivers or something like
that? Peace.

--
SCM

 
 
 

Triangle Strip Preserving LOD (T-Strip LOD)

Post by ZZ » Wed, 15 May 2002 07:49:02



Quote:> > Then my configuration is broken somewhere (regarding OpenGL atleast).

> ... which was! I was using 23.11 detonators because the nv refresh rate fix
> works with them (all fullscreen modes, where application don't allow to
> override, don't synchronize). Anyway, nVidia could address this oneday in
> their drivers - meanwhile, I guess have to either:

the problem doesnt lie with nvidia but with windows. IIRC theres a petition
aimed a microsoft to fix this problem

--
-- captain lovespunk and the goodship ecstasy -- preaching from S 4554  E
17027
http://uk.geocities.com/sloppyturds/gotterdammerung.html

 
 
 

Triangle Strip Preserving LOD (T-Strip LOD)

Post by Seppo Citymarke » Wed, 15 May 2002 15:09:57


Quote:> the problem doesnt lie with nvidia but with windows. IIRC theres a
petition
> aimed a microsoft to fix this problem

Howcome only nvidia cards for me then cause the problem? I would like to
know what makes their GeForce cards so different from my old Matrox G400
that the G400 gives picture and GeForce doesn't. But this is irrelevant as I
can always get LCD flatpanel display if I don't like CRT's. ;-)

--
SCM

 
 
 

Triangle Strip Preserving LOD (T-Strip LOD)

Post by bob » Fri, 17 May 2002 06:11:08


Interesting seeing the ping pong here about drivers, for what its worth only
have problems with nvidia drivers. If something doesn't work, dropping back
to vanilla Microsoft opengl or reducing the performance slider usually gets
things to work (excluding extensions).

Both from a development point of view and as a user something that does what
its suppose to is a lot more useful than something that has high performance
but doesn't !!!

This is a fundamental problem with all evolving technology but for those in
development and support its a nightmare. The PC came about because its a
standard not because it was good :)

Maybe Opengl needs to review the way extensions are added, with MS Direct 3d
despite the problems and bad design  of the API (compared to the simple
elegance of  opengl)  at least you know what you get with each release
.......

or perhaps Nvidia just need to employ a test engineer !

Brian.


Quote:> > the problem doesnt lie with nvidia but with windows. IIRC theres a
> petition
> > aimed a microsoft to fix this problem

> Howcome only nvidia cards for me then cause the problem? I would like to
> know what makes their GeForce cards so different from my old Matrox G400
> that the G400 gives picture and GeForce doesn't. But this is irrelevant as
I
> can always get LCD flatpanel display if I don't like CRT's. ;-)

> --
> SCM

 
 
 

1. Triangle Strip Preserving LOD (T-Strip LOD)

Hey Guys

I have a paper and demo on a new alternative to ROAM and other CLOD
algorithms for terrain LOD. I call it T-Strip LOD and it basicly subdivides
complete lines instead of triangles, esentially giving you the same
structure of a quad tree, except that blocks form a contigious line across
one axis which give you the contigious trinagle strip. It's relativly simple
and takes advantage of the high processing power of todays hardware
graphics, but with less CPU usage than conventional LOD algorithms. Perhaps
not as effective as VIPM, it may still prove to be relavent in certain
environments where there is deforming terrain, extensive terrain collision,
and mapping objects to the visble terrain mesh is required. It's more of a
cross between brute force triangle strip rendering  and CLOD.

Anyhow, the paper and demo are up at
http://chat.carleton.ca/~eszoka/tstriplod/tstrip.htm, note the demo takes 30
seconds or so to generate the terrain texture, read the readme.txt for the
keys to operate the various features and controls.

Any thoughs or comments would be appreciated.

Ernest Szoka
http://chat.carleton.ca/~eszoka/

2. Graphics 101 Question

3. Algorithm to combine triangle meshes into triangle strips

4. Computer Graphics Journals (2-5-1991)

5. Triangle List/Mesh to Triangle Strip

6. Rendering Codecs.........

7. Triangles, triangle strips, etc...

8. Research on Visual Software Products..

9. triangle mesh => triangle strip

10. Triangles vs. triangle strips

11. Problem coloring triangles using triangle strips

12. triangle mesh => triangle strip

13. Triangle Strip vs. Triangle Lists