1. Simple Quad Domains for Field Aligned Mesh Parametrization Marco Tarini Enrico Puppo Daniele Panozzo Nico Pietroni Paolo Cignoni Visual Computing Lab ISTI-CNR, Pisa, Italy Università degli Studi di Genova, Italy Università dell’Insubria Varese, Italy

2. parameterization: bijective mapping between a surface and a (planar) domain In theory… f S D f -1

3. parameterizations are extra-useful tools for: (semi-)regular remeshing texture mapping geometry processing in general physical simulations much much more… Intro f surface f -1 Some Planar Domain Good quality needed! angle + areas preserving globally smooth feature aligned transition functions across cuts Good quality needed! angle + areas preserving globally smooth feature aligned transition functions across cuts

4. Intro: good parameterizations bijectivity conformality area preservation “global smoothness” – i.e. continuity of  f u and  f v … up to rotations by K pi/4 feature alignment – alignment of  f u and  f v to geometric features “cuts” ok, if “tamed” – i.e. discontinuity of f -1 ok, if good transition functions isometry }

5. Intro: “tamed” cuts D S t Transition function: integer translation + “integer” rotation f f -1

6. Intro f surface f -1 Some Planar Domain Given this, find this and this

7. Intro: the recent breakthrough define an input cross field over S find f such that  f u,  f v match it! – (plus, enforce good transition functions across cuts) Ray, Li, Levy, Scheffer, Alliez Periodic global parametrization Kälberer, Nieser, Polthier QuadCover Bommes, Zimmer, Kobbelt Mixed Integer Quadrangulation

8. Intro: a trend in parametrization bijectivity conformality area pres. “global smoothness” feature alignment “cuts” ok, if “tamed” isometry } Ray, Li, Levy, Scheffer, Alliez Periodic global parametrization Kälberer, Nieser, Polthier QuadCover Bommes, Zimmer, Kobbelt Mixed Integer Quadrangulation make  f u,  f v match an input cross field defined on the surface constraint transition func. of f disregard (!)

9. the typical result SD f f -1 Ray, Li, Levy, Scheffer, Alliez Periodic global parametrization Kälberer, Nieser, Polthier QuadCover Bommes, Zimmer, Kobbelt Mixed Integer Quadrangulation

10. A typical U-V Domain irregular borders! tentacles! overlaps! useful tool for: (semi-)regular remeshing texture mapping geometry processing in general physical simulations much much more…

11. A typical U-V Domain Is it really ideal for semi-regular remeshing? misaligned irregular vertices (bad, for example, for reverse engeneering)

12. A typical U-V Domain Is it really ideal for semi-regular remeshing? misaligned irregular vertices (bad, for example, for reverse engeneering)

13. A typical U-V Domain irregular borders! tentacles! overlaps! useful tool for: (semi-)regular remeshing texture mapping geometry processing in general physical simulations much much more…

14. Objective: simplicity of Domain “Simplicity” of Domain – a concept encapsulating: simple shape of the domain that few cuts are used simple cut shape (e.g. lines) simple transition functions total bijectivity f surface f -1 Planar Domain

15. Objective go for Simplicity of Domain keep usual objectives too – isometry, global smoothness, alignment, good cuts keep “start from cross-field” idea

16. Similarly motivated recent work: D.Bommes, T.Lempfer, L.Kobbelt. Global structure optimization of quadrilateral meshes. CGForum 2011 A. Myles., N.Pietroni, D.Kovacs., D.Zorin Feature-aligned t-meshes. Siggraph 2010 C-H Peng, E Zhang, Y. Kobayashi, P. Wonka: Connectivity Editing for Quadrilateral Meshes SiggAsia 2011

17. Instead of irregular borders! tentacles! overlaps!

18. Proposed domain 1:1 side-to-side transitions a collection of rectangles as few as possible integer sized side-to-side adjacency (no t-junctions!) two-manifold connectivity (but no 3D embedding) useful tool for: (semi-)regular remeshing texture mapping geometry processing in general physical simulations much much more…

19. Problem statement Simple 2D domain Surface + CrossField f f -1 Given this, find this and this

20. Start from a cross field… Irregular point

21. Start from a cross field… 1.start cutting form an irregular point 2.continue cutting following the crossfield 3.end cut in another irregular point 4.repeat from 1 ==> you built a “graph of separatrices”!

22. Start from a cross field… …end up with confetti! :-(

23. Method overview 1 Input Cross-Field Initial Graph of Separatrices 1.Extract the Graph of Separatrix from the Input Cross-Field

24. Method overview 12 Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices 2.Semplify the graph of separatrices this defines a new cross field! similar to old BUT: reciprocally aligned irregular points

25. Method overview 123 Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices Global Parametrization 3.Final global optimization and smoothing includes re-tuning positions of irregular points!

26. Method overview 123 Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices Global Parametrization

27. Workflow Overview 1 Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices Global Parametrization How to do this?

28. Workflow Overview 1 Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices Global Parametrization Initial semiregular quad-mesh Discretize the cross field: build a quad resampling (use [Bommet at Al 2009] ) 0 1

29. Workflow Overview Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices Global Parametrization 1 Initial semiregular quad-mesh 0 23

30. Method overview 12 Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices Global Parametrization next: GRAPH SIMPLIFICATION ALGORITHM

31. Graph of Separatrices Irregular Point of original cross field Cross-over Separatrix following original cross field surface S q0q0 q3q3 q9q9 q2q2 q 11 q8q8 q4q4 q 10 q6q6 q5q5 q1q1 q 13 q7q7 q 12

32. Graph Simplification Opening Move ( ++ ) Continuation Move... Continuation Move Closing Move ( -- ) surface S q0q0 q3q3 q9q9 q2q2 q 11 q8q8 q4q4 q 10 q6q6 q5q5 q1q1 q 13 q7q7 q 12 one simplification cycle

33. Opening Move

35. Continuation Move

36. ? ? ok !

37. Continuation Move ok !

38. Continuation Move

39. before

40. Continuation Move after

41. Continuation Move after cross-field  (“drift”)

42. Energy associated to a separatrix total drift strive to follow original field ===> “change cross-Field the least” total lenght strive to make them short ===> “simplify graph the most” the only parameter (we used K = 5) note: it’s scale independent!

43. After many Continuation Moves…

44. Closing Move

46. Graph Simplification Opening Move ( ++ ) Continuation Move... Continuation Move Closing Move ( -- ) simplification cycle 1 Opening Move ( ++ ) Continuation Move... Continuation Move Closing Move ( -- ) simplification cycle 2...

47. Searching the solution space consistent inconsistent States: Potential moves: opening continuation closing Strategies depth first (backtracking mechanism) greedy (most appealing moves first) pruning (refuse overall energy increases) dynamic prog. (mark already seen states)

48. Method overview 12 Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices Global Parametrization GRAPH SIMPLIFICATION ALGORITHM results

50. Simplification Results

52. Method overview Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices Global Parametrization 1 Initial semiregular quad-mesh 0 23

53. 2 Workflow Overview Input Cross-Field Initial Graph of Separatrices Simplified Graph of Separatrices Global Parametrization 1 0 3 next: FINAL SMOOTHING PHASE

54. Final smoothing: objective Initial Graph Simplified Graph Final Parametrization A B misaligned singularitiesaligned singularitiesre-optimized AA A BB B graph simplification final smoothing

55. Final Optimization Phase: summary Responsible for 1.relax param inside patches 2.smooth lines between patches 3.optimize positions of singularities ! How to: – “quincux” strategy – exploit domain simplicity!

56. Results 4524 rectangular regions 98 rectangular regions

57. Results 1368 rectangular regions 82 rectangular regions

58. Results 33973 rectangular regions 124 rectangular regions

59. Results 498 rectangular regions 112 rectangular regions

60. Results 2387 rectangular regions 150 rectangular regions

61. Summary Concept: Domain simplicity is important for usability of parameterizations! Our main result: Drastic simplifications of domains typically achieved The catch: Reliance on quality of input cross field. E.g.: superfluous/too-misplaced singularities ==> grief

62. What about crease lines? don’t? preserve? recover them while smoothing

63. Interpolation domains interpolation domains parameterization domain

64. Interpolating points over the domain parameterization domain

65. Workflow in practice 2 Initial semiregular quad-mesh Initial Separatrices Graph Simplified Graph Global Parametrization Input Cross-Field 1 0 3 Discretize the cross field: build a quad resampling (use [Bommet at Al 2009] )

66. Workflow in practice 2 Initial semiregular quad-mesh Initial Separatrices Graph Simplified Graph Global Parametrization Input Cross-Field 1 0 3

67. Workflow in practice 2 Initial semiregular quad-mesh Initial Separatrices Graph Global Parametrization Input Cross-Field 1 0 3 Simplified Graph