meshingddg.dreamhosters.com/cs1762014/wp-content/uploads/2014/01/meshi… · meshing of mri...

6
1 Making Meshes and Putting Them To Good Use Making Meshes and Putting Them To Good Use Mathieu Desbrun Caltech CS176 31 So What Is Wrong in 3D? Regular tets do NOT tile space perfectly shaped tets everywhere? dream on DT of well-spaced points can have (lotta) slivers assuming "well-spaced" is well defined… 32 Scrutinizing the Beast Four (well-spaced) vertices near the equatorial plane of their circumsphere 33 Taming the Beast? Recipes to get rid of slivers local jiggling of faulty vertices through optimization of shape’s “quality” for example, penalize small dihedral angles no guarantee of success—in fact, may become worse locally altering the Delaunay triangulation called “sliver exudation” no longer Delaunay, obviously but weighted Delaunay—local tweak of the metric » will be discussed later guarantees, but no practical bounds let’s go one dimension higher 34 Lifting Transformation (nD) Lift points (x, y) to paraboloid (x, y, x 2 + y 2 ) lift points up; compute convex hull what is the induced connectivity? Delaunay triang. = convex hull of 3D points 35 Lifting Transformation (nD) Lift points (x, y) to paraboloid (x, y, x 2 + y 2 ) what about Voronoi diagrams? Voronoi diagram = proj. of upper envelope Before…

Upload: others

Post on 03-Feb-2021

0 views

Category:

Documents


0 download

TRANSCRIPT

  • 1

    Making Meshesand Putting Them To Good Use

    Making Meshesand Putting Them To Good Use

    Mathieu DesbrunCaltech

    CS176

    31

    So What Is Wrong in 3D?

    Regular tets do NOT tile space perfectly shaped tets everywhere? dream on DT of well-spaced points can have (lotta) slivers

    assuming "well-spaced" is well defined…

    32

    Scrutinizing the Beast

    Four (well-spaced) vertices near the equatorial plane of their circumsphere

    33

    Taming the Beast?

    Recipes to get rid of slivers local jiggling of faulty vertices

    through optimization of shape’s “quality” for example, penalize small dihedral angles

    no guarantee of success—in fact, may become worse

    locally altering the Delaunay triangulation called “sliver exudation” no longer Delaunay, obviously

    but weighted Delaunay—local tweak of the metric» will be discussed later

    guarantees, but no practical bounds let’s go one dimension higher

    34

    Lifting Transformation (nD)

    Lift points (x, y) to paraboloid (x, y, x2 + y2) lift points up; compute convex hull what is the induced connectivity?

    Delaunay triang. = convex hull of 3D points

    35

    Lifting Transformation (nD)

    Lift points (x, y) to paraboloid (x, y, x2 + y2) what about Voronoi diagrams?

    Voronoi diagram = proj. of upper envelope

    Before…

  • 2

    36

    Pushing This Analogy Further

    Idea: move vertices to perfect the approximation of the paraboloid roots in function approximation

    Hessian(x2/2) = Id studied quite a bit in 2D Lloyd algo:

    for given points,VD gives optimal L1 approx

    for given conn., find optimal vertex positions always move a vertex to the centroid of its Voronoi region Live Demo

    volume btw approximant and paraboloid

    37

    Lloyd Algorithm for Meshingread input boundary setup data structure & preprocessinggenerate initial sites inside do

    - Delaunay triangulation of {xi}- move sites to optimal locations {xi*}

    until convergence or stopping criterionextract interior mesh

    38

    (after Lloyd relaxation)

    Breaks down, unfortunately…

    why? Voronoi cells seem well-shaped but plenty of poorly-shaped tets, as mentioned… use of a dual approach, maybe?

    Lloyd Algorithm in 3D

    39

    Under/Overlaid Approximant

    CVT PWL approximant compact Voronoi cells isotropic sampling

    ODT [Chen 04] PWL interpolant compact simplices isotropic meshing

    40

    Optimization of Vertices

    Energy minimum for given connectivity iff:

    Massage the equation and you get:

    ij ikjk xxxkij

    ii xxx

    ,

    2* ||||)(||||2

    1

    j xi

    xk

    ij

    jji

    i cx

    ||||

    1*

    circumcenter of j

    41

    Results of Optimization in 2D

    Note: needs proper boundary conditions…

    Demo

  • 3

    42

    0

    Optimization

    Distribution of radius ratios

    43

    Optimization

    1

    44

    Optimization

    2

    45

    Optimization

    46

    Optimization

    47

    Optimization

  • 4

    48

    The Visible Human

    Meshing of MRI datasets

    49

    Other Big Meshes

    Dihedral angles ≥ 15.0

    50

    Fandisk (Mechanical Part)

    51

    Comparisons

    Dihedral angles ≥ 3°4,189 vertices

    TetGen

    Dihedral angles ≥ 0.1°15,157 vertices

    DelPSCNODT

    Dihedral angles ≥ 12°4,050 vertices

    52

    Power Diagrams

    Offseting squared distances by weights

    Voronoi: all points treated equal

    now “weight” wi per vertex

    weights alter the L2 distance

    covers set of orthogonal duals (V-1 DOFs)Simple and efficient generalization surprising applications as well

    0.4

    [Aurenhammer et al. ‘98]

    53

    Meshes with power diagrams

    Can find a dual diagram by tweaking the weights

  • 5

    54

    Adaptive Sampling (who needs meshing :)

    55

    Architecture via power diagrams

    Discrete Force field within the walls forms cells

    56

    Putting Meshes to Work

    Once meshing done, computations can happen Eulerian approaches

    static mesh, motion encoded through local cell exchanges common for fluid simulations, level sets

    57

    Smoking Bunny

    58

    Smoke without Aliasing

    59

    Liquid Shape

    Fluid with free surface (using foliation processing)

  • 6

    60

    Putting Meshes to Work

    Once meshing done, computations can happen Eulerian approaches

    static mesh, motion encoded through local cell exchanges common for fluid simulations, level sets

    Lagrangian approaches mesh moves (initial mesh = rest position) common for elastic and plastic deformable bodies

    61

    Meshes that Deform

    Elastic deformations of gargoyles and armadillos

    62

    Conclusions

    Meshing: fascinating research field needs theoretical guarantees

    with practical bounds

    needs practical results with theoretical foundations

    can be extremely complex think quad/hexa meshes... plenty of remaining challenges

    small angles, accelerating convergence, handling of anisotropy

    Now it’s your turn!