Bungie Halo2 illustration of dot3 bump map modeling

[BenSkywalker]

As in, you have a base texture map, a specular / diffuse, lightmap.. etc. All of these are going to be combined (multi-textured) before the final values arrive. Why don't they combine these textures into one and bake them into ONE texture beforehand instead of doing the combining at the rendering stage?

Think of a floor with a texture on it. Let's say it is a brick floor. Taking a look at your typical game if you see one brick floor, you are likely to see quite a few of them. Not only that, but there will likely be a repeating texture to make even a single example of a brick floor.

Now think of the lighting model. Say it is night time and there are lights dispersed throughout the area. Using the current standard you would simply apply the differing lightmaps over the area they are used in. If you wanted to revert to a single texture you would need a unique texture for each section of the floor, exploding the amount of memory utilized by what is currently a relatively small handful of textures.

For using the trick for things like diffuse lighting, damage etc, your possibilities for a single texture could reach into the millions- for each and every texture in the game. Think of firing a rocket across the area with the brick floor. For each frame you would need to load a different texture for each area effected by the light given off from the rocket. Then, whichever areas it hit would need to load another set of textures to demonstrate the damage and then another set of textures still would need to be used for the next time you fired off a rocket as you would need the differing lighting model due to the rocket along with the leaving the damge visible. For a singular room you could chew up TBs worth of data trying to compute out all of the possible changes a small handful of textures go through.

 

[JF_Aidan_Pryde]

Thanks Ben.

I realise the need for multitexturing when it's affected by lighting but on the previous page they were discussing such details such as the dents on Master Chief's armour and veins in the mosters of Doom3. Such things were not induced by lighting / damage but were there in the first place. No lighting is going to change the fact that monsters in doom will have veins. I don't see any case for veins to be turned on / off from external factors. Also the imperfections on the armour of Master Chief were there in the first place, so I take it that no multitexturing would need to be done? A true damage map, granted would need multitexturing, but the original scratches should be pre-baked, yes?

 

[cybamerc]

Tagrineth:

> True, but also as I said the normal maps lose their effect at extreme
> angles, the end result being a great big waste.

A waste? What kind of nonsense is that.


JF_Aidan_Pryde:

> Why don't they combine these textures into one and bake them into
> ONE texture beforehand instead of doing the combining at the rendering
> stage?

Because they have different purposes and different attributes. Color maps, normal maps and gloss maps for example look nothing alike. Even grayscale textures such as light maps, gloss maps and height maps can't (in general) be combined as the resulting effect depends not only on the math done on the gfx hardware but on the specific look of the map.

As Ben also mentions memory plays a huge factor.

> they were discussing such details such as the dents on Master Chief's
> armour and veins in the mosters of Doom3. Such things were not
> induced by lighting / damage but were there in the first place.

The dents and scrathes are actually included in the normal map. It only makes sense too as you we're talking about bumps in the surface.

As for the monsters in Doom 3 the veins are painted on color map but larger veins may also be included in the normal map to make them look bumpy.

> but the original scratches should be pre-baked, yes?

Finer details such as scratches can be painted directly on the color map but high lights and defined shadows should not be included when you have bump maps, specular maps, gloss maps etc.

 

[JF_Aidan_Pryde]

Thanks cybamerc, that pretty much clears it up.

 

[V3]

I was saying 5000 poly is still low poly for a complex model, though alot higher than the 300 or so poly of last gen.

 

[malcolm]

what i dont understand is why doom models have cube fingers if they have 5000 polys?
5000 realy seems anough to make things round.

 

[Ozymandis]

what i dont understand is why doom models have cube fingers if they have 5000 polys?
5000 realy seems anough to make things round.

I thought I read somewhere that the models were between 2000 and 5000 depending on the particular model (and perhaps they have different detail levels for models like in Quake3?).

 

[randycat99]

The "cube finger"- sounds like a medical condition! Along with other common afflictions such as block face, pointy shoulder, facet fore-arm...

 

[V3]

what i dont understand is why doom models have cube fingers if they have 5000 polys?
5000 realy seems anough to make things round.

Its not enough.

 

[BoddoZerg]

I was saying 5000 poly is still low poly for a complex model, though alot higher than the 300 or so poly of last gen.

When was the last time models were 300 polies? Quake 2? Thats a bit more than one generation ago... heck, even models in strategy games (Warcraft III) have 500 polies or so.

Just how many polies are in a UT2k3 model anyways? It's definitely WAY higher than Halo, Halo2, or Doom3.

 

[Tagrineth]

Tagrineth:

> True, but also as I said the normal maps lose their effect at extreme
> angles, the end result being a great big waste.

A waste? What kind of nonsense is that.

I didn't mean they were wastes overall, I mean obviously bump mapping is pretty much mandatory these days for a realistic image.

My point was that if you're looking at a surface with a normal map at an extreme angle, those bump maps really gain an artificial look, and no longer look bumpy at all, so at those angles, real polygons would look infinitely better.

 

[randycat99]

I was told that only the emboss mapping technique had this problem at extreme angles, but you are saying good ole DOT3 has the same problem, too?

 

[Tagrineth]

I was told that only the emboss mapping technique had this problem at extreme angles, but you are saying good ole DOT3 has the same problem, too?

Emboss's problem is when the light is at an extreme angle.

I'm talking about the viewport being at an extreme angle... EVERY bump mapping form breaks down with that condition.

 

[randycat99]

Interesting!

So I guess in the end, there is no replacement for real polys? ...Or is it possible for someone to come up with a newer bumpmapping scheme down the line that remains intact even under extreme angles?

So what sort of artifact is actually seen at the extreme angles? Does the bump cease to be bump, or do you see some sort of geometry floating above the parent model? What?

 

[aaaaa00]

So what sort of artifact is actually seen at the extreme angles? Does the bump cease to be bump, or do you see some sort of geometry floating above the parent model? What?

The bump starts to look flat instead of bumpy.

 

[V3]

When was the last time models were 300 polies? Quake 2? Thats a bit more than one generation ago... heck, even models in strategy games (Warcraft III) have 500 polies or so.

Err we are in console forum, last gen was PSX, N64 and Saturn.

 

[randycat99]

Heh-heh, you got da cubefinger!

 

[Laa-Yosh]

Some reasons to use multitexturing...
- specular hihglights are dynamic, so the specular/gloss maps can not be baked
- dirt, scratch and other maps can be re-used for different surfaces
- detail maps can have a different tiling on the surface, resulting in a better texel/pixel ratio when viewed up close
- decals (blood stains etc.) are obvious

 

[Simon F]

So what sort of artifact is actually seen at the extreme angles? Does the bump cease to be bump, or do you see some sort of geometry floating above the parent model? What?

The bump starts to look flat instead of bumpy.

It's somewhat ironic, don't you think, when you consider that the models are generally polygonal and so otherwise smooth-looking surfaces look bumpy at the edges

 

[Brimstone]

Interesting!

So I guess in the end, there is no replacement for real polys? ...Or is it possible for someone to come up with a newer bumpmapping scheme down the line that remains intact even under extreme angles?

So what sort of artifact is actually seen at the extreme angles? Does the bump cease to be bump, or do you see some sort of geometry floating above the parent model? What?

I think displacement mapping is considered the next step beyond bump mapping. Displacement mapping tesselates and deforms polys as needed so it is more of a polygon type of solution, at least thats how I understand it.