基本双轮廓理论

发布于 2024-11-17 09:16:46 字数 120 浏览 5 评论 0原文

我一直在谷歌上搜索,但找不到任何基本的东西。在最基本的形式中,双轮廓(对于体素地形)是如何实现的?我知道它的作用以及原因,但不明白如何做到这一点。 JS 或 C#(最好)都不错。以前有人使用过 Dual轮廓并能简单解释一下吗?

I've been searching on google, but cannot find anything basic. In it's most basic form, how is dual contouring (for a voxel terrain) implememted? I know what it does, and why, but cannot understand how to do it. JS or C# (preferably) is good.Has anyone used Dual contouring before and can explain it briefly?

如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。

扫码二维码加入Web技术交流群

发布评论

需要 登录 才能够评论, 你可以免费 注册 一个本站的账号。

评论(1

我恋#小黄人 2024-11-24 09:16:46

好的。所以今晚我很无聊,决定自己尝试一下双重轮廓。正如我在评论中所说,所有相关材料均位于以下论文的第 2 节中:

特别是,部分描述了网格的拓扑2.2 在下面的部分中,引用:

  1. 对于每个表现出符号变化的立方体,生成一个位于方程 1 的二次函数的最小值处的顶点。

  2. 对于每条表现出符号变化的边,生成一个连接包含该边的四个立方体的最小化顶点的四边形。

这就是全部了!您解决线性最小二乘问题以获得每个立方体的顶点,然后用四边形连接相邻顶点。因此,利用这个基本思想,我使用 numpy 在 python 中编写了一个双轮廓等值面提取器(部分只是为了满足我自己对其工作原理的病态好奇心)。代码如下:

import numpy as np
import numpy.linalg as la
import scipy.optimize as opt
import itertools as it

#Cardinal directions
dirs = [ [1,0,0], [0,1,0], [0,0,1] ]

#Vertices of cube
cube_verts = [ np.array([x, y, z])
    for x in range(2)
    for y in range(2)
    for z in range(2) ]

#Edges of cube
cube_edges = [ 
    [ k for (k,v) in enumerate(cube_verts) if v[i] == a and v[j] == b ]
    for a in range(2)
    for b in range(2)
    for i in range(3) 
    for j in range(3) if i != j ]

#Use non-linear root finding to compute intersection point
def estimate_hermite(f, df, v0, v1):
    t0 = opt.brentq(lambda t : f((1.-t)*v0 + t*v1), 0, 1)
    x0 = (1.-t0)*v0 + t0*v1
    return (x0, df(x0))

#Input:
# f = implicit function
# df = gradient of f
# nc = resolution
def dual_contour(f, df, nc):

    #Compute vertices
    dc_verts = []
    vindex   = {}
    for x,y,z in it.product(range(nc), range(nc), range(nc)):
        o = np.array([x,y,z])

        #Get signs for cube
        cube_signs = [ f(o+v)>0 for v in cube_verts ]

        if all(cube_signs) or not any(cube_signs):
            continue

        #Estimate hermite data
        h_data = [ estimate_hermite(f, df, o+cube_verts[e[0]], o+cube_verts[e[1]]) 
            for e in cube_edges if cube_signs[e[0]] != cube_signs[e[1]] ]

        #Solve qef to get vertex
        A = [ n for p,n in h_data ]
        b = [ np.dot(p,n) for p,n in h_data ]
        v, residue, rank, s = la.lstsq(A, b)

        #Throw out failed solutions
        if la.norm(v-o) > 2:
            continue

        #Emit one vertex per every cube that crosses
        vindex[ tuple(o) ] = len(dc_verts)
        dc_verts.append(v)

    #Construct faces
    dc_faces = []
    for x,y,z in it.product(range(nc), range(nc), range(nc)):
        if not (x,y,z) in vindex:
            continue

        #Emit one face per each edge that crosses
        o = np.array([x,y,z])   
        for i in range(3):
            for j in range(i):
                if tuple(o + dirs[i]) in vindex and tuple(o + dirs[j]) in vindex and tuple(o + dirs[i] + dirs[j]) in vindex:
                    dc_faces.append( [vindex[tuple(o)], vindex[tuple(o+dirs[i])], vindex[tuple(o+dirs[j])]] )
                    dc_faces.append( [vindex[tuple(o+dirs[i]+dirs[j])], vindex[tuple(o+dirs[j])], vindex[tuple(o+dirs[i])]] )

    return dc_verts, dc_faces

它不是很快,因为它使用 SciPy 的通用非线性求根方法来查找等值面上的边缘点。然而,它似乎确实工作得相当好并且以相当通用的方式。为了测试它,我使用以下测试用例(在 Mayavi2 可视化工具包中)运行它:

import enthought.mayavi.mlab as mlab

center = np.array([16,16,16])
radius = 10

def test_f(x):
    d = x-center
    return np.dot(d,d) - radius**2

def test_df(x):
    d = x-center
    return d / sqrt(np.dot(d,d))

verts, tris = dual_contour(f, df, n)

mlab.triangular_mesh( 
            [ v[0] for v in verts ],
            [ v[1] for v in verts ],
            [ v[2] for v in verts ],
            tris)

这将球体定义为隐式方程,并通过双轮廓求解 0-等值面。当我在工具包中运行它时,结果如下:

在此处输入图像描述

总之,它似乎有效。

Ok. So I got bored tonight and decided to give implementing dual contouring myself a shot. Like I said in the comments, all the relevant material is in section 2 of the following paper:

In particular, the topology of the mesh is described in part 2.2 in the following section, quote:

  1. For each cube that exhibits a sign change, generate a vertex positioned at the minimizer of the quadratic function of equation 1.

  2. For each edge that exhibits a sign change, generate a quad connecting the minimizing vertices of the four cubes containing the edge.

That's all there is to it! You solve a linear least squares problem to get a vertex for each cube, then you connect adjacent vertices with quads. So using this basic idea, I wrote a dual contouring isosurface extractor in python using numpy (partly just to satisfy my own morbid curiosity on how it worked). Here is the code:

import numpy as np
import numpy.linalg as la
import scipy.optimize as opt
import itertools as it

#Cardinal directions
dirs = [ [1,0,0], [0,1,0], [0,0,1] ]

#Vertices of cube
cube_verts = [ np.array([x, y, z])
    for x in range(2)
    for y in range(2)
    for z in range(2) ]

#Edges of cube
cube_edges = [ 
    [ k for (k,v) in enumerate(cube_verts) if v[i] == a and v[j] == b ]
    for a in range(2)
    for b in range(2)
    for i in range(3) 
    for j in range(3) if i != j ]

#Use non-linear root finding to compute intersection point
def estimate_hermite(f, df, v0, v1):
    t0 = opt.brentq(lambda t : f((1.-t)*v0 + t*v1), 0, 1)
    x0 = (1.-t0)*v0 + t0*v1
    return (x0, df(x0))

#Input:
# f = implicit function
# df = gradient of f
# nc = resolution
def dual_contour(f, df, nc):

    #Compute vertices
    dc_verts = []
    vindex   = {}
    for x,y,z in it.product(range(nc), range(nc), range(nc)):
        o = np.array([x,y,z])

        #Get signs for cube
        cube_signs = [ f(o+v)>0 for v in cube_verts ]

        if all(cube_signs) or not any(cube_signs):
            continue

        #Estimate hermite data
        h_data = [ estimate_hermite(f, df, o+cube_verts[e[0]], o+cube_verts[e[1]]) 
            for e in cube_edges if cube_signs[e[0]] != cube_signs[e[1]] ]

        #Solve qef to get vertex
        A = [ n for p,n in h_data ]
        b = [ np.dot(p,n) for p,n in h_data ]
        v, residue, rank, s = la.lstsq(A, b)

        #Throw out failed solutions
        if la.norm(v-o) > 2:
            continue

        #Emit one vertex per every cube that crosses
        vindex[ tuple(o) ] = len(dc_verts)
        dc_verts.append(v)

    #Construct faces
    dc_faces = []
    for x,y,z in it.product(range(nc), range(nc), range(nc)):
        if not (x,y,z) in vindex:
            continue

        #Emit one face per each edge that crosses
        o = np.array([x,y,z])   
        for i in range(3):
            for j in range(i):
                if tuple(o + dirs[i]) in vindex and tuple(o + dirs[j]) in vindex and tuple(o + dirs[i] + dirs[j]) in vindex:
                    dc_faces.append( [vindex[tuple(o)], vindex[tuple(o+dirs[i])], vindex[tuple(o+dirs[j])]] )
                    dc_faces.append( [vindex[tuple(o+dirs[i]+dirs[j])], vindex[tuple(o+dirs[j])], vindex[tuple(o+dirs[i])]] )

    return dc_verts, dc_faces

It is not very fast because it uses the SciPy's generic non-linear root finding methods to find the edge points on the isosurface. However, it does seem to work reasonably well and in a fairly generic way. To test it, I ran it using the following test case (in the Mayavi2 visualization toolkit):

import enthought.mayavi.mlab as mlab

center = np.array([16,16,16])
radius = 10

def test_f(x):
    d = x-center
    return np.dot(d,d) - radius**2

def test_df(x):
    d = x-center
    return d / sqrt(np.dot(d,d))

verts, tris = dual_contour(f, df, n)

mlab.triangular_mesh( 
            [ v[0] for v in verts ],
            [ v[1] for v in verts ],
            [ v[2] for v in verts ],
            tris)

This defines a sphere as an implicit equation, and solves for the 0-isosurface by dual contouring. When I ran it in the toolkit, this was the result:

enter image description here

In conclusion, it appears to be working.

~没有更多了~
我们使用 Cookies 和其他技术来定制您的体验包括您的登录状态等。通过阅读我们的 隐私政策 了解更多相关信息。 单击 接受 或继续使用网站,即表示您同意使用 Cookies 和您的相关数据。
原文