The Australian Game Developers Conference

[mckmas8808]

Not really. Normal Maps simply present the illusion of more geometry. If you had at your disposal enough polys to model with, you wouldn't need normal maps and other such tricks. Indeed, having the actual geometry there if you were able to would be more desireable, I think.

Of course, we often don't have the capability often to throw around as many polys as we'd like, or perhaps as cheaply as something like normal mapping can appear to, hence its valuable role.

If your source model is say 1 million polys, does your in-game model have to be less than 10,000 polys?

Can you make the in-game model 100,000 polys if you wanted too?

 

[jvd]

Not really. Normal Maps simply present the illusion of more geometry. If you had at your disposal enough polys to model with, you wouldn't need normal maps and other such tricks. Indeed, having the actual geometry there if you were able to would be more desireable, I think.

Of course, we often don't have the capability often to throw around as many polys as we'd like, or perhaps as cheaply as something like normal mapping can appear to, hence its valuable role.

Of course both is better than either .

I rather have an increase in poylgon counts and an increase in normal map space (texture ram / compresion) than eitherj ust polygon counts or normal map space

a 100 million polygon scene will look much better with normal maps than without

 

[jvd]

If your source model is say 1 million polys, does your in-game model have to be less than 10,000 polys?

Can you make the in-game model 100,000 polys if you wanted too?

you can do whatever you want . THe point is to get that 1 million polygon look for alot less polygons .

Look at the first minute or so of that video. They talk about taking the backround from that scene which was 200 million polygons and converting it to a scene that is 1 million polygons but with normal maps .

That is a huge amount of power saved that can be used for other things .

 

[mckmas8808]

you can do whatever you want . THe point is to get that 1 million polygon look for alot less polygons .

Look at the first minute or so of that video. They talk about taking the backround from that scene which was 200 million polygons and converting it to a scene that is 1 million polygons but with normal maps .

That is a huge amount of power saved that can be used for other things .

So that means that Lair could be used normal maps. Hmmm.....

 

[jvd]

So that means that Lair could be used normal maps. Hmmm.....

what system is lair on ? It may not be able to .

If its a ps2 game i doubt they have the texture room to use good normal maps


But look at it this way mckmas .

Say a system can draw max 2 billion polygons at once. That would look way better than gears of war . However . How about if you then take 10 billion polygon models and use normal maps to make them a few hundred million polygons .

The image quality on that would look much better than the original 2 billion polygon scene . Of course u need the texture space for it

 

[mckmas8808]

what system is lair on ? It may not be able to .

If its a ps2 game i doubt they have the texture room to use good normal maps


But look at it this way mckmas .

Say a system can draw max 2 billion polygons at once. That would look way better than gears of war . However . How about if you then take 10 billion polygon models and use normal maps to make them a few hundred million polygons .

The image quality on that would look much better than the original 2 billion polygon scene . Of course u need the texture space for it

Hey that actually sounds correct. Wow I like normal maps now. Well I probably always liked them and didn't know they were being used.

Oh and Lair is being made on the PS3.

 

[Bobbler]

Normal maps are merely a trick to fake geometry. If we had no limit in the polys we could push then normal maps would serve no purpose. It's another trick added to the countless others to fake reality, since calculating the real world accurately takes all the elementary particles in existance all working together.

a 10million raw poly model would probably be preferable to a 10k poly model + 20million poly normal map -- but normal mapping a 10k poly model is vastly cheaper and looks good enough. Normal mapping suffers from some issues (hence the reason for the push for things like parallax mapping) -- look at the heads of the humans in Doom 3, those models are normal mapped a ton but the silhouette looks like a mess.

 

[jvd]

Oh and Lair is being made on the PS3.

then its most likely already using normal maps

I hope the ps3 uses 3Dc for normal map compresion . Hell i wish they both use 3Dc +

 

[Titanio]

Of course both is better than either .

I rather have an increase in poylgon counts and an increase in normal map space (texture ram / compresion) than eitherj ust polygon counts or normal map space

a 100 million polygon scene will look much better with normal maps than without

Well, you may get to a point of diminishing returns. If you're talking about hundreds of millions of polys, well, you're possibly getting down into levels of detail that your resolution might have difficulty keeping up with anyway. If you can model all the detail you want given the geometry you have, and without things like normal maps, all the better. You're only using things like normal maps in the absence of sufficient polys to model every detail you want to show.

My point is, given enough geometry, you don't need to fall back on texture tricks. All these techniques do is provide the illusion of more geometry. So it doesn't make much sense to say that normal maps provide "a much bigger leap" than polys can, given that all these techniques are doing is mimicking more polys.

 

[jvd]

Well, you may get to a point of diminishing returns. And if you can model all the detail you want given the geometry you have, and without things like normal maps, so be it.

My point is, given enough geometry, you don't need to fall back on texture tricks. All these techniques do is provide the illusion of more geometry. So it doesn't make much sense to say that normal maps provide "a much bigger leap" than polys can, given that all these techniques are mimicking is more polys.

I'm pretty sure that texture room has increased much more from last gen to this gen than polygon counts have . Actual usable in game polygon counts

This will most likely continue into the future as ram costs continue to drop and modual sizes increase .

Aside from that till we hit the point were we see diminishing returns we are going to want to fake as much as possible and i believe that point is way off into the future

 

[Titanio]

I'm pretty sure that texture room has increased much more from last gen to this gen than polygon counts have . Actual usable in game polygon counts

This will most likely continue into the future as ram costs continue to drop and modual sizes increase .

Aside from that till we hit the point were we see diminishing returns we are going to want to fake as much as possible and i believe that point is way off into the future

I agree more or less, but that's a seperate point. I can see texture trickery being attractive for some time yet, absolutely. I'm just posing the point that if you could use enough polys, there'd be no need for these techniques.

The question Bobbler poses about high poly counts versus lower counts + normal maps derived from even higher poly counts than the former is an interesting one though, that may become more relevant. As you say, possibly having both would be the ideal.

 

[zidane1strife]

Well, you may get to a point of diminishing returns. If you're talking about hundreds of millions of polys, well, you're possibly getting down into levels of detail that your resolution might have difficulty keeping up with anyway. If you can model all the detail you want given the geometry you have, and without things like normal maps, all the better. You're only using things like normal maps in the absence of sufficient polys to model every detail you want to show.

My point is, given enough geometry, you don't need to fall back on texture tricks. All these techniques do is provide the illusion of more geometry. So it doesn't make much sense to say that normal maps provide "a much bigger leap" than polys can, given that all these techniques are doing is mimicking more polys.

Indeed a multi-billion poly scene probably has as much detail as one can discern on any modern display, if it doesn't reach the limits of what humans can perceive. I think the problem is also animating the stuff. From what I hear even if you could render a static object with high lvls of geometry, with all that added geometry it'd be tough to process it for say a human character, though I may be wrong I'm going of from my fuzzy memory.

From what I've seen of normal mapping, it is a very good technique and should be used widely in the enviroments, and for many possible clothing accesories(like body armor), also for adding smaller details that would probably be more cumbersome to handle with geometry. But for things like loose clothing, it's good to have high-geometry so as to properly simulate the clothes movement, also for things like doing proper animated skin creases(see snake's neck/face when he moves and his skin folds accurately) and better animating facial features, etc .

 

[ERP]

I agree, but that's a seperate point. I can see texture trickery being attractive for some time yet, absolutely. I'm just posing the point that if you could use enough polys, there'd be no need for these techniques.

Unfortunately this is untrue....
Textures have nice properties in terms of antialiasing that do not exist with very detailed geometry. Very high res geometry has all sort of aliasing artifacts that cannot be trivially fixed.

Imagine a window sill on a building far enough away that the window sill is less that one pixel (or sample) in width, that window sill can be brightly lit with respect to the building face, and that leads to a sparkling bright set of pixels moving along the line. Short of some form of perfect analytical AA even with very large amounts of geometry practical you'll still want to do something in your LOD system like

High Res Mesh -> Lower Res Mesh + Normal maps -> Low Res mesh + BRDF

And there are places in the High res mesh where normal maps are actually still preferable to geometry detail.

 

[zidane1strife]

The GT4 shot was a photo in the scenery, and Laa-yosh already suggested that as a solution for Motorstorm, though obviously what we're seeing looks a lot more 3D than a 2D photo, and hence Laa-Yosh's scepticism. I don't think looking at a photo backdrop is any indicator of what Motorstorm can or can't do with 3D terrain.

U can drive through that portion of the background(the road goes there, from what I've heard.). It's supposedly low geometry with a photo texture on top, I don't've first hand experience with that track so I'm not able to verify if that's the case or not, but IIRC that's what I heard.

Yeah they said that the monsters were only 7 to 9 thousand polygons. How is that possible? Well I know how it's possible because the video told me how, but will these monsters look worst than the dragons in Lair? Not to say the monsters in Gears of War look bad because they look well, "Unreal".

And I'm only asking because the dragons in Lair were supposed to be over 100,000 polygons, so don't think this is a systems war question.

The monsters in lair are using normal mapping for smaller details and static details probably, that higher poly count is what probably allows the dragons to look like their skin/scales are moving over some muscles, and as if some muscles are bulging underneath, and also for superb sphere like and round like features such as the eye and the eye-lid along with the highly moving-organic-fleshy-look all over, and for things like the wings(which seem to have some sort of smooth cloth'esque animation or physics). Those flexible realistically moving/flexing thin wings are probably part of the reason for the uber geometry lvls.

edit: look at the HIGH rez lair video over at kikizo(maybe in the same link I provided in the ps3 gundam over at the console games section). You can see that unlike other models these models don't've the 'static' look they look malleable and their muscles seemingly move and bulge under the skin, and their wings flap and look smooth throughout the flat thin surface, with realistic cloth'esque animation, along with superb organic spherical features.

 

[Titanio]

Unfortunately this is untrue....
Textures have nice properties in terms of antialiasing that do not exist with very detailed geometry. Very high res geometry has all sort of aliasing artifacts that cannot be trivially fixed.

Imagine a window sill on a building far enough away that the window sill is less that one pixel (or sample) in width, that window sill can be brightly lit with respect to the building face, and that leads to a sparkling bright set of pixels moving along the line. Short of some form of perfect analytical AA even with very large amounts of geometry practical you'll still want to do something in your LOD system like

High Res Mesh -> Lower Res Mesh + Normal maps -> Low Res mesh + BRDF

And there are places in the High res mesh where normal maps are actually still preferable to geometry detail.

Thanks! While I was typing I was trying to think "is there some other reason to use mapping techniques?" and just nothing came to me in the context of a limitless poly budget, since I was thinking only in about modelling concerns But that makes sense.

 

[Bobbler]

Imagine a window sill on a building far enough away that the window sill is less that one pixel (or sample) in width, that window sill can be brightly lit with respect to the building face, and that leads to a sparkling bright set of pixels moving along the line. Short of some form of perfect analytical AA even with very large amounts of geometry practical you'll still want to do something in your LOD system like

Isn't that more of an issue with resolution, rather than inherent issues with geometry? If we're dreaming here... we might as well take into account a display with resolution far higher than what we have now then I don't think you see these aliasing issues. In today's world the various texturing techniques solve problems because of inherent limitations in our hardware (displays/resolution and gpu's/bandwidth/etc).

Maybe there's something else to it, but it seems to me that without the restrictions we currently have geometry would always be preferable to a texturing trick doing the same job (if we want perfect picture quality, that is).

 

[mckmas8808]

UThe monsters in lair are using normal mapping for smaller details and static details probably, that higher poly count is what probably allows the dragons to look like their skin/scales are moving over some muscles, and as if some muscles are bulging underneath, and also for superb sphere like and round like features such as the eye and the eye-lid along with the highly moving-organic-fleshy-look all over, and for things like the wings(which seem to have some sort of smooth cloth'esque animation or physics). Those flexible realistically moving/flexing thin wings are probably part of the reason for the uber geometry lvls.

edit: look at the HIGH rez lair video over at kikizo(maybe in the same link I provided in the ps3 gundam over at the console games section). You can see that unlike other models these models don't've the 'static' look they look malleable and their muscles seemingly move and bulge under the skin, and their wings flap and look smooth throughout the flat thin surface, with realistic cloth'esque animation, along with superb organic spherical features.

You are a big help. So basically if a dev was to create a animal, alien, human, etc. and wanted to have a high level of animation/the stuff you explained then using a higher number of polys is what the dev should use?

Well that question may be a bit confusing. Let me ask this. (Not picking on GOW or the Xbox 360) What is it that the monsters in Gears of War would lack (being that Epic took the Lower Res Mesh + Normal maps route) that another dev using a high number of polys would have?

 

[Laa-Yosh]

What is it that the monsters in Gears of War would lack (being that Epic took the Lower Res Mesh + Normal maps route) that another dev using a high number of polys would have?

Silhouette detail, self-shadowing detail, and deformation accuracy (for both the body muscle/skin movement and facial animation). You also shouldn't expect the chains on that big monster to do any kind of motion (ie. waggle around his neck).
Up close, texture filtering artifacts would also pop up, but then again polygonal detail wouldn't look that better in such a close-up either, only differently ugly.

 

[Laa-Yosh]

Looks like everyone got distracted by my comment on that hill (which I'll get back to in a minute) and missed the more important point. So I'll repeat.

Every single techdemo and finished game that we've seen from the nextgen consoles - that have been proven to be real and realtime - share some very important common qualities. From PGR3 and Kameo, through Gears of War and MGS4 and Heavenly Sword and Fight Night, to Lair and Project Offset, they are all relying on the combination of implementations and extensions of well-known hardware rendering techniques, different mostly in art direction and efficiency. Many of them are justly considered to be state of the art, coming from the leading studios of the industry - but they're all using texture mappedd polygons, normal mapping and (in CG terms) simple surface shaders; they're all suffering from various geometry and shading aliasing artifacts, and they're all still easily identified as realtime rendered scenes. Especially the image quality (the one developers can do the least about, given the limits of the hardware) is noticeably different from the polished look of most CGI, but this is also the first thing that's cured in a shakycam video.

My point is that we're still just taking the first steps into this next generation of consoles, and developers are still in the stages of adapting to the new technology. While I'm somewhat aware of the exciting new possibilities of the hardware, experimentation is still constrained by budgets and deadlines. It is unrealistic to expect developers to totally revolutionize hardware rendering techniques in such a short amount of time, and it is foolish to expect or even demand CG-like quality from the upcoming games. People should have learned by now, based on the above mentioned games' graphics, what they should get to see; and it should also be at least a bit more obvious when a publisher, developer or hardware manufacturer tries to fool us with prerendered stuff. Unfortunately, it seems that they are still able to do it again and again... and I wonder if it's worth arguing about, or should I just wait and let time decide it.


So, getting back at those hills, I've said that if it is not a photo (or otherwise pre-calculated and baked) texture, than it's not likely to be realtime. As good as normal mapping is, it'd still not be enough to produce that kind of self-shadowing and soft lighting on those hills, you need geometry to do that, and loads of it. But please take your time and read my post again, it was just the first thing I've mentioned because there's a lot in that small video already that suggests it to be CGI...

 

[mckmas8808]

Silhouette detail, self-shadowing detail, and deformation accuracy (for both the body muscle/skin movement and facial animation). You also shouldn't expect the chains on that big monster to do any kind of motion (ie. waggle around his neck).
Up close, texture filtering artifacts would also pop up, but then again polygonal detail wouldn't look that better in such a close-up either, only differently ugly.

I've seen that word (Silhouette detail) before. What exactly does it mean? I understand the rest of your post and it makes sense.