@El'endiaStarman Yes sir?

@El'endiaStarman I implemented both marching cubes and squares, do you need help in it?

I'm trying to figure out a good way to take a collection of cubes (Minecraft-style) and turn them into smoother, polygonal surfaces.

Yeah, first thing you need to learn when doing a voxel-based project, that voxels aren't cubes

Minecraft isn't a voxel game, it's a voxel based game

You need to start with a 3dimensional array of values

@El'endiaStarman

@Bálint I do have one. They're just all 0 or 1, so it seems like the marching cubes algorithm doesn't quite apply.

If you want to get a blocky result, you need to use 1 or 0, if you want a more smooth mesh, then you need to have values between 0 and 1

@El'endiaStarman I assume you use a pseudo-random noise generator, don't you?

@El'endiaStarman We should create a separate discussion for this though, so it doesn't get mixed with other stuff

@Bálint I have an idea for a 3D FPS where you fly around in spaceships, and I want to have an environment that's basically 3D Perlin/simplex noise. If I implement a noise generator, I can definitely use marching cubes. However, I already have a biased majority rule implementation that produces fairly nice terrain (you saw me post a picture of it). So I'm trying to figure out a good way to go from that to nice-ish surfaces.

I don't want high-resolution surfaces, necessarily. Just something better than cubes.

I should note that the language I'm using is Blitz 2D/3D, which is basically a game engine, and correspondingly less powerful in terms of more-abstract data structures and the like, though it's great for graphics and interactivity.

@wizzwizz4 That wouldn't help me make meshes that approximates the cube structures.

Incidentally, @Bálint, do you know Blender?

@El'endiaStarman Wait, I create a separate discussion for this

So

Hang on, lemme move the messages over.

Your current algorithm can only generate 1 or 0 as values, or can it do it inbetween them

?

Also, mind if I change this room's parent site to PPCG?

No, I don't

Huh, looks like that's not possible. Oh well. [shrugs]

Back to the topic: Your current algorithm can only generate 1 or 0 as values, or can it do it inbetween them?

← 19 messages moved from The Nineteenth Byte

@Bálint It's only 0 or 1 at present. Do you know the majority rule cellular automata?

I do

You can solve that, if you want a mesh with slopes at a 45 degrees angle

The marching cubes algorithm posted at paulborke.net contains a vertex interpolation part

Wich makes the values actually correspond to the shape of the mesh, and not just the isolevel

You want a 3d planear regression algorithm.

You can remove this, and then you should get a look close to this:

@El'endiaStarman

Hmm, that does look fairly close to what I want.

Can you post your code somwhere?

I could put it on GitHub, yeah.

Then please do it

@El'endiaStarman I found a better image

The first is where no marching cubes is applied

The second is marching cubes without interpolation

Oooh, (b) is basically exactly what I was thinking of!

And the third is with interpolation

Yeah

Where did you get the algorithm from?

For my terrain generation? It's just a biased form of 3D majority rule. That's what I'll put on GitHub.

No, the marching cubes

Oh, I don't have anything for that yet.

I understand

Then I don't need anything

Have you read the article at paulborke.net?

@El'endiaStarman This one: paulbourke.net/geometry/polygonise

Yes, I have.

I'm unsure how to apply it though.

First, copy and paste the lookup tables, namly the edge table and the tritable

I can't really do that though.

Why?

This is an example of the language I'm working with.

Besides, I have a binary field. Can I not use something simpler than marching cubes?

Marching cubes is the simplest

@El'endiaStarman The other probability is dual contouring, but that's really hard

But do you have arrays?

I do have arrays. Either local and 1D or global and N-D.

And arrays can't contain arrays?

Nope.

That's going to be hard

Yeah.

What does N-D mean?

N-dimensional.

Then you could use that, don't you?

I could. I still wouldn't be able to just copy-paste though.

What is the format of it? I can write a throwaway program for converting

I could too.

I'm just trying to be lazy. :P

Being lazy is when you create a program, to do the work for you

hmm

Just tell me how an array looks like, and I do it

Hmm. Blitz doesn't have array literals like most languages. There is a way to do it though, but it's a feature I've never used.

But really, is there literally no simpler way to do it?

I mean, this is another example of the terrain:

No

Due to the way that majority rule CAs work, there are fewer general arrangements of cubes.

Sadly

@El'endiaStarman

You could make test cases, so if there's a block below and on the left, create a slope between them

@El'endiaStarman But this would require a huge amount of cases

I did this manually in Blender, but is there any reason this couldn't be easily automated without requiring huge lookup tables?

I told you above, you can do it with test cases

Surely there's a better way.

But that's a lot of case

@El'endiaStarman There's 256 of them

2 of wich you don't insert a new triangle

I'm fairly sure this is different from how marching cubes would do it.

Sketch of an algorithm: for each empty space, look for these patterns [...] and where they exist, add a new facet in the appropriate way.

I've got some progress - pairs of adjacent centers now get that angled facet.