PDA

View Full Version : Real Sub-pixel Displacement



kfinla
10-24-2005, 01:07 AM
i know lots of ppl have been asking for micro-poly displacement etc in the past. I just dont want that request to die. I think i read about some sort of fake sub-pixel displacement hack being added to Lw9.

Anyways it would be really nice if the LW renderer adaptively tesselated, and just handeled displacements better in general. Being able to work with the new 32 bit normal and displacemnt maps coming outta Zbrush would sure be nice.

I hate how much better displacements seem to be handled in most of the other render engine, and how many have REAL micro, or sub-pixel displacement.

"While not true micro-poly displacement, this method will yield similar results in many cases." - from LW9 feature list.

Captain Obvious
10-24-2005, 01:29 AM
The adaptive subdivision in LW9 isn't fake and it isn't a hack. Adaptive tesselation is what it is, so I don't see the big problem. It's not quite the same thing as micropoly displacement, but it will look the same.

kfinla
10-24-2005, 01:43 AM
I guess im both wondering how this is different, and then why it can't be "true micro-poly displacement". I'm not getting the vibe their saying LW9's method is superior, its sounds like a work around. I get the vibe, meh its close. Thats why i used the word hack. I mean PRman is hands down the best engine out there at handling displacements, cuz pixar owns the Reyes algorithm as well as lots of other things. Its just many packages have this feature, even poser.

I mean its cool i guess, that there your gonna be able to choose how things tesselate on a calculation by calculation basis. Maybe weird (bad) actually if the renderer is looking at differenet geometry for each calc. I just want the ability to do the real thing if i want, not just tune how good the fake is. The ability to optimize is great if it can be 100% accurate also.

Captain Obvious
10-24-2005, 02:43 AM
As far as I know, adaptive subdivision has a couple of advantages and a couple of disadvantages, like most things. I'm pretty sure adaptive subdivision is how Pixels 3D's Tempest renderer does it, and it works really well. In fact, you don't need to use bump-maps at all, and the render hit for using it is next to nothing.

Sure, PRMan is better at it (fine detail displacement) regardless, but Lightwave has better raytracing and global illumination, and better price and infinite render nodes and and and... It's a trade-off, like most things.

Mylenium
10-24-2005, 04:30 AM
I guess im both wondering how this is different, and then why it can't be "true micro-poly displacement". I'm not getting the vibe their saying LW9's method is superior, its sounds like a work around. I get the vibe, meh its close. Thats why i used the word hack. I mean PRman is hands down the best engine out there at handling displacements, cuz pixar owns the Reyes algorithm as well as lots of other things. Its just many packages have this feature, even poser.

I mean its cool i guess, that there your gonna be able to choose how things tesselate on a calculation by calculation basis. Maybe weird (bad) actually if the renderer is looking at differenet geometry for each calc. I just want the ability to do the real thing if i want, not just tune how good the fake is. The ability to optimize is great if it can be 100% accurate also.

I fail to see your point. Even micropolygon tesselation needs to be tweaked and technically it's entirely possible it uses a different tesselation for each frame. It can look just as crap as LW's coming adaptive subdivision if not adjusted properly. The only problem I see is that adaptive subdivision will use more memory, but that's all. So really, assuming it is a "fake" or workaround is far from the truth.

Mylenium

kfinla
10-24-2005, 07:07 AM
Well I guess unfortunatly i'm complaining about somthing that isn't out yet. I just expect to be disappointed, so hopefully ill be pleasantly surprised. I'm just sick of rendering in Zbrush cuz LW's displacement is so poor.

It would also be nice if LW updated the "normal displacement" plugin. Lynx 3d's version is far better than NT's, but only allows up to 10 instances of it. If your breaking apart characters this is no good. Turtle for instance lets you have 99.

Also have LW recognize and be cool with 16 bit grayscale, and now 32! bit displacment maps. Again Lynx 3d to the rescue (not sure about the 32, but dont think i could tell anyways, since I dont think the renderer currently would make use of the data or be able to accurately represent it)

If LW wants to be a serious package its gonna have to be able to deal with more than the 4-6 million polys it currently craps out at (on my machine atleast with 2gb of ram). Ppl are doing work in the 30 million range on characters. Hopefully Lw64 will help.

BtW - im not talking about different geometry calcs per frame being potantially weird, im talking about the renderer looking at different rez's of geometry for reflection, displacement, GI, refraction.. i just think it could be messy. Thats the way i read the statement.

Lightwolf
10-24-2005, 07:21 AM
Again Lynx 3d to the rescue (not sure about the 32, but dont think i could tell anyways, since I dont think the renderer currently would make use of the data or be able to accurately represent it)

Well, LW internally either uses 8bit integer per image channel, or 32bit float, so it would surely use a 32bit float image if that existed. Actually lynxs 16bit TIFF loader loads those TIFFs into a 32bit float image anyhow.

Cheers,
Mike

Sculley
10-24-2005, 09:14 AM
I cant think why NT's tesselated method can be called fake... even PRMans has limitations.. such as bounding limits.. how far it can be pushed before you dont even see your geometry.. but then.. if you look at the PRMan specs.. you realize everything about it is designed to *fake* everything quickly at film quality.. take for instance.. Pixars shadows dont work with motion blur. Just cos Pixar do something.. dont think its perfect! Not to mention.. it took Pixar a long time to fix there tears between patches/buckets.

In PRMan you can displace only with in a limit.. which is frankly really small... NT's method should allow for huge displacements. I'm looking forward to it.

Captain Obvious
10-24-2005, 01:50 PM
BtW - im not talking about different geometry calcs per frame being potantially weird, im talking about the renderer looking at different rez's of geometry for reflection, displacement, GI, refraction.. i just think it could be messy. Thats the way i read the statement.
As far as I've been able to figure it out, Lightwave 9 will use the same exact tesselation for all things in a single frame. That means no raytracing problems.

kfinla
10-24-2005, 03:28 PM
Separate tessellation values for shading, shadow and reflection/refraction channels

"The user can set separate level of detail settings for shading, shadows and reflection tessellation of the sub-division surface..."

From the Lw9 feature list media handout.

Could be cool, could be wacky.

loki74
10-24-2005, 06:43 PM
Correct me if I'm wrong, but aren't a LOT of the render things we do actually "fake?" I mean, the very fact that you're looking at an image rendered to RGB values could be called "fake." The only renderer I know of that does not do this is maxwell. Lens flares--aren't those "fake" as well? Or how about the very nature of 3D modeling--as I understand it, all points are infinitely small, and all polygons are infinitely thin. If we were to be literally "real," would we not be building all of our models... molecule by molecule? Dynamic simulations, CFD, these could be called fake as well. Even RealFlow isn't 100% accurate CFD, although it comes pretty damm close.

Lets face it--there are too many variables, too many things going on in real life to have anything be "100% accurate." What matters is getting the render to be accurate enough to achieve the desired degree of realism or style, but no more accurate than this to save render times.

Isn't this the whole premise of adaptive subdivision? It won't highly subdivide things that are so far away or out of view, because a satisfactory degree of accuracy in the final product can be achieved without the extra detal, and time can be saved.

kfinla
10-24-2005, 10:06 PM
Believe it or not im looking quite forward to LW9, i really wasnt sure i was gonna bother with it but re-reading the feature update it does look like by LW 11 or so, LW will essentially be completely re-written, which its needed to be since version 7.5?

Displacements in LW currently are pretty crappy. So im glad this is being addressed. I just look at every update of LW as a chance to surpasse the competition not just play catch up or be almost as good as the current state of the art. All im getting at is a lot of rendering engines have real micropoly displacement, so why cant we.

Sculley
10-25-2005, 08:41 AM
Micro-poly displacement is there in apps like Renderman because of how it operates.. it dices geometry in to teeny-weeny geometry.. and shades it.. once that is done.. and only when.. the engine then moves its minute polygons (though its really quads.. remembering renderman doesnt like triangles)... it then has to re-calculate the shaders... which is pretty wastefull...

An engine like LightWave doesnt work in the same way.

from what I understand.. NT are displacing geomentry and constantly dicing it till its at the smallest required size.. i.e a pixel. Though im not sure if it will work to a subpixel level... im pretty sure you'ld not notice any difference.. you will stil lget the same level of detail even to the fact you'll be able to displace far beyond most other engines limitations.

Most micro-poly displacement are simply called because geometry is diced to the smallest viewed size.. a poly that covers a single pixel... im pretty sure that NT's implementation will be making jsut as small polys to get the result. So in the long run I dopnt think there will be any difference... apart from the ability to apply displacements like hills etc and not see it dissappear in the render engine.