I've been trying to understand how to generate volumetric data in order to render voxel terrain.

I'm trying to seperate the generation into parts, and understand the various components. I'm making one assumption at the beginning and if I'm wrong at the onset, please let me know.

I'm assuming that creating a voxel engine requires you to do a few steps:

1. Populate volumetric data by generating noise and filling the data structure with what will be rendered

2. The volumetric data is converted into polygons using an algorithm like marching cubes

3. The vertex buffers generated are draw to the screen

I'm starting with #1 here, and I'm finding some problems understanding how to use NOISE to generate 3D volumetric data that will look realistic. I'm also having problems understanding how one generates an entire scene like this... Assuming that the scene is seperated into "CHUNKS" which are drawn. I'm not sure I understand how each chunk would know about the other chunk, in order to make the transitions smooth. As you generate more chunks to keep the scene going indefinitely. How does a single noise function tell us everything about the world? how do we know where the SKY is? How do we know where the DIRT is? Why is there a seperation between the dirt, and the ground? how is the inside of the ground "solid? while the sky is "empty"? I understand about density when talking about height maps, but how is this same concept applied to the air, and the dirt inside the ground?

Of course, when you think Heightmap, it's all very simple to understand to some degree (although you still have the chunk problem) but it only defines the "surface" of the terrain. Since the Voxel terrain works differently, how do you then use Noise to generate volumetric data that will look like real terrain when it has to include everything that is INSIDE the chunk as well, not just where the "ground" is. If i'm thinking about it "minecraft" style, it's a little bit easier to understand, because of the way the cubes would work. Noise could work quite well in this method, and I have generated this in the past, but using smooth voxel terrain isnt quite the same to me when it comes to noise and how you would generate actual terrain.

I guess I'm missing a piece of the puzzle here that I'm sure would clear things up in my mind.

Having said all this. If I just go and read reference material, like how perlin noise works, etc which I have, it doesn't explain any of the above questions. Thats why I'm hoping that people will help me understand this concept specifically within the context of my question, not in general terms about "how perlin noise works" if that makes sense.

Any help is appreciated. i'll probably bounce off ideas back and forth if someone is willing to help me out!

3Ddreamer, I guess you highlighted the very problem in your answer. There are plenty of posts, but none actually explain. They just talk about it as if you already understand the subject matter. It makes it difficult to get a handle on it when you can't find any actual explanations that give you the answers you need.

Right now, I'm not interested in the problem you describe about games using this or that, thats not my concern. I'm just trying to learn and understand how to do this for my brain to know, not to produce a video game in the end specifically. I'll use all my knowledge to do that in the end. I need answers to specifics right now.

You pointed out using Noise, but you didn't give me answers as to HOW. You are just pointing me to more questions. (although I appreciate you replying!)

I've read almost every thread in gamedev.net that has to do with Voxel Terrain and Noise. All of them are discussions among people who already have understanding of what I'm missing. So noone points it out because everyone understands it. So I'm left behind because I need specific answers.

At this point, I'm totally ok with having a boring terrain. I just want to generate actual terrain. I've seen some articles that talk about using noise TOGETHER to generate the terrain. The part they skip every time, is how are you putting 2 noises together? ;P

There are several things I do not understand about what I've seen so far:

1. No mention of how you "merge" multiple noise function results together. People talk about it, about using different methods "together" to generate more natural looking terrain. But they don't go into HOW you do that. Are we saying that you multiply the values? divide? add?

2. No mention of how you use noise, to generate volumetric data, even at it's most basic concept. Like when you talk about Voxel Engines, and you talk about a chunk, being a three dimensional array of density values, how do the values on the top end up being sky? and how do the values on the bottom end up being ground?

When you are "filling" the volumetric data structure, you do:

for (int x = 0; x <= size; x++)
{
    for (int y = 0; y <= size; y++)
    {
        for (int z = 0; z <= size; z++)
        {
            double div = 64.0;
            double val = (SimplexNoise.noise((x) / div, (y) / div, (z) / div)) * 128.0;

            VolumeData[x, y, z] = (sbyte)val;
        }
    }
}

But this doesn't generate terrain, it generates a sponge square. So i'm just confused at this point, how 1 function call to SimplexNoise or PerlinNoise, etc. can generate volumetric data that will resemble "terrain". Especially if you are not using a "minecraft" cube system, but using more of a marching cubes type of smooth terrain. (Although should that matter? isn't the volumetric data the same, just represented differently by different types of algorithms when building the polygons?)

.

3Ddreamer, telling me to go back to school isn't very helpful... I'd say that's not what most Indie developers do to learn something new.

Anyway, thanks to everyone else who responded with really important details. I think at least i knew what I didn't know (or what was missing in my head). It's the fact that I'm getting basic noise because thats what I'm generating :P I need to use Noise properly in order to generate the volumetric data.

Once I have the volumetric data matrix for the region i want to draw, then I have to go to step 2, which is to generate polygons for the volumetric data that I generated in order to render it to the screen.

I did not realize that the noise used to generate basic minecraft like volumetric data, would be very similar to the noise used to generate more complex volumetric information. I suppose it's all in the resolution of the data. While minecraft data would be stored as on/off for whether a specific cube in a chunk exists or not, the more complex volumetric data, would still contain these "regions" or "chunks" but instead, they would describe a single cube or so, of actual space. So now, instead of having lets say:

World -> Region -> Cube (On/Off)

so lets say a region is 32x32x32, and the world is 32x32x32 regions, then you get 1024x1024x1024 cubes.

While in the latter, you probably have something more like:

World -> Region -> Section (Matrix [x,y,z] of densities?)

So now, you could still process the world using the Octree but when you get down to the "cube" or "section" level, you are not just on/off, you are another smaller matrix of volumetric data for that cube region?

Am I understanding the basic data structure? (if what i said even made sense to people, at least it's the words that come to my mind)

My planet renderer uses a ridged multifractal generator (8 octaves of Simplex noise) added to a fractal brownian noise generator (12 octaves of Simplex noise), plus an additional two layers of Simplex noise for each of the generators to perturb their input parameters. That's 24 calls to a Simplex noise function for every pixel of the terrain. And I'm working in 2D - for good looking voxel terrain, I'd probably be using significantly more layers of noise.

Of course that's amazing! I have seen some amazing results. I'm sure that you progressed over time through various discoveries as you created this planet render (which sounds amazing BTW).

At this point I have much much smaller goals. Even if I could generate something as simple as basic rolling hills with a sky vs. ground, that would make me happy and at the very least would allow me to move forward with additional concepts.

At least now I have the ability to focus my energies in generating 3D volumetric data with Noise functions. Of course, now it's a matter of figuring out how to do more of the things you mentioned. I've seen SOME breakdowns in different places but they are usually focusing their attention on other things, and never really break down the basics. Once of them did break down basics, but it was written in this Lua which I have no idea about yet. So I wasn't able to get much detail from it.

JTippets, thanks for that answer. It goes a bit beyond what I'm trying to make sense of right now, however I appreciate the details.

I do not have a well rounded mathematical education. This I cannot change. I have to work with what I have, in order to achieve my goal, there's simply no other way. I cannot go back to university in order to do what I"m trying to do... I have to learn on my own with what i"ve got.

Having said that, I understand generally what you are saying, and it's making sense to me in a general sense. Once I get past the current problem I am having, I believe that playing with these functions in some type of pre-made terrain generator, may assist me in figuring out what types of noise functions can produce various end results.

The part where you talk about the Sky/Ground and how we are talking simply about values of density between -1 and 1 then in terms of starting with a normal 2D heighmap makes sense to me.

Although even that, is tripping me up right now. I know that the volume data is a Matrix of values [x,y,z] The marching cubes algorithm takes this data, and then uses it to generate the mesh.

If the volume data was all 1's or all -1's should this produce a CUBE then for that region of "space"? and a volume data matrix of all values of zero, would be blank space?

If you remove the noise entirely, shouldn't you be able to lets say fill the matrix with half 1's and half 0's and get Half a cube?

I want to make sure I understand the volume data and what it's meant to represent, just to make sure I didn't have a faulty premise on what the volume data is suppose to represent.