This repository was archived by the owner on Sep 1, 2020. It is now read-only.
Repository files navigation 200819_DigitalGeometryProcessing
曲面上任何函数的平滑都可以用拉普拉斯
原始问题难以求解时,构建中间变量,最后再进行转换
划分成低频和高频,高频在局部坐标系(n,t1,t2),低频下的操作可以达到实时,高频通过局部坐标来映射
Local/Global 迭代两个优化变量(普遍收敛较慢)
网格表示:
点云
符号距离场
隐函数
Grid 网格:八叉树,
Mesh 网格:半边结构
离散微分几何:
局部平均区域
法向量
梯度
拉普拉斯算子
曲率
平均曲率:(max+min)/2
高斯曲率:max x min
网格平滑:
filter-based:基于滤波(滤波算子,迭代算子)
Laplacian Smooth:Laplacian operator
Gaussian Denoise
Bilateral Denoise
Bilateral Normal Filter
Fourier Transform:manifold harmonics(流形谐波)
optimization-based:基于数值优化算法
Paper: Image smoothing via L0 gradient minmization
-> Mesh demoising
Differential Edge Laplacian Operator
Paper: Mesh denoising vai L0 minimization
Paper: Variational Mesh Denoising using Total Variation and Piecewise Constant Function Space
先优化法向量,再计算顶点的位置
以法向量作为中间变量,通过转化的方式来求解
data-driven:基于神经网络
Paper: Mesh Denoising via Cascaded Normal Regression
网格参数化(parameterization、flattening、unfolding):
Constraints
Bijective
Inversion-free
Locally injective
Low distortion
Tutte’s barycentric mapping:开网格 -> 圆盘
find boundary(easily with halfedge)
boundary to circle
solve sparse matrix
Least squares conformal maps(LSCM, ASAP)
Conformal map:包角变换,局部是相似变换
LSCM:目标函数为使映射与相似映射的差值最小
确定两个点:
避免平凡解(0)
去除旋转和平移带来的不唯一性
Angle-Based Flattening (ABF:保角变换)
先计算角度,再由角度计算顶点的参数化坐标
计算角度:no Linear ABF -> Linear ABF
重构参数化:Reconstruct parameterization
Greed method:数值误差的累积
Least squares method:误差会均匀分布,误差会扩散到每一条边,避免了误差的累积
以角度作为中间变量,通过转化的方式来求解 *
As-rigid-as-possible (ARAP:尽量 isometric)
Paper: A Local/Global Approach to Mesh Parameterization
尽量 isometric(只有平移加选择的变换)
优化雅克比和旋转矩阵差的 F 范式最小
交替迭代
固定雅克比矩阵,优化旋转矩阵(Local):signed SVG
固定旋转矩阵,优化雅可比矩阵(Global):最小二乘,正交方程
Paper: Computing Inversion-Free Mappings by Simplex Assembly
Edge assembly constraints:确保三角形可以拼起来
网格变形(Deformation):
Definition
Fixed + Deformation Region + User Handle
Transformation Propagation
转换扩散:handle 到 fixed 光滑插值(smooth blending)
计算测地距离 or 内部拉普拉斯为零,边界为 1(迭代 or solve)
Multi-Scale Deformation
划分成低频和高频,在低频进行 deform 后叠加高频
高频用局部坐标系来存储(n,t1,t2)
Differential Coordinates
Gradient-Based Deformation
Laplacian-Based Deformation
变换插值
Deformation transfer:将一个网格的变形转换到其他网格(两个网格存在一一映射)
三角形不能确定唯一的仿射变换,增加一个 v1 + n 顶点
encode transfer(encoder-decoder?)
As-Rigid-As-Possible surface deformation
优化的角度,所有的变换都是旋转
交替迭代(Local/Global)
热启动(每次迭代以上次的迭代结果为迭代初值)
Freeform Deformation • Meshless mapping
mesh deform -> space deform
Lattice-Based Freeform Deformation
Cage-Based Freeform Deformation
外部变换传递到内部:重心坐标(变换前后的 GBC 不变)
Volumetric Deformation • Tetrahedral mapping
总结
handle-based deformation
cage-based deformation
skeleton-based deformation
重心坐标(Barycentric):
Introduction
非负性
求和为1
reproduction,可以由顶点的坐标组合构造出所有的内点
Barycentric coordinates on convex polygons
Generalized barycentric coordinates(GBC)
形式和双线性插值相同
Inverse bilinear coordinates
Mean value coordinates(MVC)
MVC 公式
Motivation of MVC:逼近 harmonic maps(调和映射,拉普拉斯为零)
Harmonic Coordinates
A general construction
Skeleton-subspace deformation
Paper: Bounded Biharmonic Weights for Real-Time Deformation
Points-handles:局部微调变换
Bones-handles:局部刚性变换
Cages-handles:整体变换
有局部性,局部变换不会扩散到其他顶点,不会有负值和反转
网格修复(Repair)(介绍为主 ):
Definitions
Application dependent(only display or FEM)
Artifacts 产生
激光扫描产生
Non-uniform rational B-spline(NURBS) -> Marching Cube -> Mesh 产生
PointCloud -> Mesh 产生
Defects and flaws
孤立点和悬挂边
奇异边
奇异点
拓扑噪声(虚假环炳)
法线方向(面朝向)不一致
补洞(具有主观性,模糊性,不确定性)
狭缝(分裂或重叠)
退化三角形
自相交(穿模)
锋利的倒角
数据噪声
Upstream and Downstream applications
根据 Upstream 确定 Artifacts 类型
根据 Downstream 确定需要修复到什么程度
Types of input
Registered range scans(注册范围扫描)
Fused range scans(融合范围扫描)
Triangle soups(一坨三角形)
Triangulated NURBS patches(多个 NURB 单独三角化后连接到一起)
Contoured meshes(体素提取获得的网络)
Badly meshed manifolds(质量差的三角形网格)
Approaches
Volumetric algorithms(体素算法:Mesh -> grid)
映射(Mapping):
Introduction
Mappings
Mesh-based mapping
Meshless mapping
Applications
Parameterization
Deformation
Mesh Improvement
Compatible remeshing(surface mapping)
Basic requirements(Foldover-free: det(J)>0)
Basic goal (As Rigid As Possible, As similar as Possible)
Distortion of J is about its eigenvalue: min D(f)
Maintenance-based methods
Paper: Most-Isometric ParameterizationS (MIPS,满足约束的不好解 -> 满足约束的好解)
初始有效映射
随机选择顶点
牛顿迭代
重复 2-3
输出有效映射
Bounded distortion methods
Paper: Bounded distortion mapping spaces for triangular meshes
Representation-based method
Paper: Computing inversion-free mappings by simplex assembly
多立方体生成(PolyCube)(介绍为主 ):
Definition
Tetrahedral Mesh -> PolyCube -> All-Hex Mesh
Deformation-based method
All-Hex Mesh Generation via Volumetric PolyCube Deformation
Voxel-based method
Optimizing PolyCube domain construction for hexahedral remeshing
compute polycube
compute mesh-polycube mapping
Cluster-based method
PolyCut: Monotone Graph-Cuts for PolyCube Base-Complex Construction
Generalized PolyCube
表面映射(Surface Mapping)(介绍为主 ):
Definition
one-to-one mapping between surfaces
Application
Morphing(变形)
Attribute transfer
Algorithms
Common base domain
A -> C <- B(都转化到一个简单的网格表面)
Parameterization-based method
A -> C <- B(都转化到一个边界相同的参数域)
cutting path
compute theta, phi
bijection lifting
形状插值(Morphing, shape interpolation):
Definition
(Source, Target) -> interpolation
Angle, length, area, volume, and curvature
Example-Driven Deformations Based on Discrete Shells
基于无限小的采样
Affine transformation
As-Rigid-As-Possible Shape Interpolation
基于仿射变换的插值
简单插值
SVD 插值
RS 插值(对极分解:rotation & scaler)(效果最好 )
Data-driven morphing(no mention)
A Data-Driven Approach to Realistic Shape Morphing
Data-Driven Shape Interpolation and Morphing Editing
点云注册/对齐(Point set registration)
ICP:两步迭代
寻找最近的对应点
计算误差最小的刚体变换
迭代 1-2
纹理映射生成(Atlas generation):
Definition
Requires defining a mapping from the model space to the texture space
Mesh cutting
Point -> Paths
Segmention
Paper: D-Charts: Quasi-Developable Mesh Segmentation
Chart parameterization
Bijective, low distortion
Scaffold(脚手架)
Paper: Simplicial Complex Augmentation Framework for Bijective Maps
Chart packing(略)
Paper: Atlas Refinement with Bounded Packing Efficiency
Packing Efficiency
Low distortion
Consistent orientation
Overlap free
Low boundary length
Atlas Refinement Pipeline
Overloap Paramterization ->
Axis-Aligned Structure ->
Distortion Reduction
网格简化(Simplification):
Definition
曲率保持 or 曲率移除(保持高曲率的点,或移除高曲率的点)
Local operations
点移除
边坍缩
半边坍缩(会产生非法的拓扑结构)
Quadric error metric(QEM)
计算每个点的二次误差度量
每个边有一个二次误差度量
每次迭代移除误差最小的边
重新计算涉及到的边的二次误差度量
Variational shape approximation
Paper: Variational shape approximation (VSA)
用大面片替换若干小面片
球面参数化:
Definition & Applications
Hierarchical method
Paper: Advanced Hierarchical Spherical Parameterizations
Flat-to-extrusive decimation strategy:先移除接近平面,再移除高曲率平面
triangle mesh
Decimation 简化,去掉高曲率
Curvature error metric (CEM)
Refinement 精炼,按照原来的顺序逆序将点加回来
triangular sphere
Two hemispheres
先映射到两个圆盘上,再映射到两个半球上(用球极投影融合,或者直接融合)
Curvature flow(ricci flow)
方向场(Directional Field):
Introduction
magnitude & direction
Vector fields
Cross fields
Frame fields
Discretization
切空间:
面的切空间是法向量
点和边的切空间是加权平均的法向量
index:转一圈的角度差除以 2pi,不等于 0 的为奇异点
Representation
局部坐标系 + Angle-based = 1-direction field
局部坐标系 + Angle-based-N-Symmetry = N-symmetry-dicection field
Objectives and Constraints
Objectives:Fairness(光滑)
Constraints:Alignment(对齐)
可积场是无旋的
Paper: Frame Fields: Anisotropic and Non-Orthogonal Cross Fields
重新网格化:
Introduction
网格质量提升且与原网格接近
网格质量提升修复的是非拓扑的问题
优化:采样密度,三角形正规化
保持:Element alignment and orientation
Isotropic remeshing
Incremental Remeshing:
Edge collapse
Edge split
Edge filp
Vertex reclocation
Project to surface(CGAL: AABB tree)
Error-bounded method: Hausdorff distance
Anisotropic remeshing
Delaunay 三角化:
Introduction
Convex hull:凸包
Carathéodory’s Theorem:通过计算凸包的点来计算凸包,O(N^4)
The Separation Theorem:通过计算凸包的边来计算凸包,O(N^3)
Triangulation
Delaunay triangulation
Delaunay Triangulation(空圆性:任何三角形的外接圆内部不包含其他点)
The Lawson Flip algorithm
初始化三角化
不满足空圆性的边进行 flip
Properties
Empty Circle:空圆性
Maximize the minimum angle:最大化最小角
Optimal Delaunay triangulation
改变顶点位置,使得最小角进一步最大化
优化 flip map 的三菱锥和抛物面的体积差
Voronoi 图:
Introduction
Post Office Problem
Voronoi Diagram
Voronoi Cell:每个区域里面距离凸多边形的点最近
Duality
Voronoi 图 和 Delaunay 三角化,是相互对偶的关系
Centroidal Voronoi tessellations (CVT)
Definition
Applications
Algorithms
网格表示:两点最短路径 & 最小生成树
离散微分几何:可视化平均曲率和高斯曲率(梯度)
网格平滑:Paper: Bilateral Normal Filtering for Mesh Denoising
迭代计算新的面 normal
恢复顶点位置
保持三角网格的体积
网格参数化:Tutte’s barycentric mapping
开网格 -> 圆盘(uniform laplacian)
网格参数化:Paper: A Local/Global Approach to Mesh Parameterization
网格变形:Paper: As-Rigid-As-Possible surface deformation
重心坐标:Mean value coordinates(Tutte’s barycentric mapping
开网格 -> 圆盘(MVC laplacian))
形状插值:基于仿射变换的 RS 插值
网格简化:边坍缩算法
方向场:复数多项式的形式优化 Cross fields,四边形网格生成
重新网格化:Incremental Remeshing(CGAL: AABB tree)
Delaunay 三角化:The Lawson Flip algorithm + Optimal Delaunay triangulation
Voronoi 图:Lloyd iteration 在网格简化上的实现:Variational shape approximation (VSA)
网格平滑:Paper: Mesh Denoising via Cascaded Normal Regression
About
No description, website, or topics provided.
Resources
Stars
Watchers
Forks
You can’t perform that action at this time.