Multidimensional Adaptive Sampling and Reconstruction for Ray - - PowerPoint PPT Presentation

▶
multidimensional adaptive sampling and reconstruction for
SMART_READER_LITE
LIVE PREVIEW

Multidimensional Adaptive Sampling and Reconstruction for Ray - - PowerPoint PPT Presentation

Jan 26, 2023 354 likes •1.3k views

Multidimensional Adaptive Sampling and Reconstruction for Ray Tracing Toshiya Hachisuka * Wojciech Jarosz * Richard Peter Weistroffer Kevin Dale Greg Humphreys Matthias Zwicker * Henrik Wann Jensen * * University of California, San

slide-1
SLIDE 1

Multidimensional Adaptive Sampling and Reconstruction for Ray Tracing

Toshiya Hachisuka* Wojciech Jarosz* Richard Peter Weistroffer† Kevin Dale‡ Greg Humphreys† Matthias Zwicker* Henrik Wann Jensen*

*University of California, San Diego †University of Virginia ‡Harvard University

slide-2
SLIDE 2
slide-3
SLIDE 3
slide-4
SLIDE 4

Sampling and Realistic Rendering

  • Distributed ray tracing [Cook et al. 84]

4

Image Plane Lens Objects Pixel

slide-5
SLIDE 5
slide-6
SLIDE 6
slide-7
SLIDE 7
slide-8
SLIDE 8

Adaptive Sampling

  • Recursive subdivision [Whitted 80]
  • Stochastic adaptive sampling [Mitchell 87, 91]
  • Perceptually based criterion [Bolin and Meyer 98]
  • Divergence based criterion [Rigau et al. 03]
  • MISER [Press and Farrar 90] [Leeson 03]

8

Image based adaptive sampling

slide-9
SLIDE 9

Adaptive Sampling

  • Recursive subdivision [Whitted 80]
  • Stochastic adaptive sampling [Mitchell 87, 91]
  • Perceptually based criterion [Bolin and Meyer 98]
  • Divergence based criterion [Rigau et al. 03]
  • MISER [Press and Farrar 90] [Leeson 03]

9

Image based adaptive sampling

slide-10
SLIDE 10

Motion Blur

10

slide-11
SLIDE 11

Motion Blur

11

slide-12
SLIDE 12

Motion Blur

12

slide-13
SLIDE 13

Motion Blur

Time

13

slide-14
SLIDE 14

Motion Blur

Time Pixel

13

slide-15
SLIDE 15

Motion Blur

Time Pixel

14

slide-16
SLIDE 16

Motion Blur

Time Pixel

14

slide-17
SLIDE 17

Motion Blur

Time Pixel

15

slide-18
SLIDE 18

Motion Blur

Time Pixel

16

slide-19
SLIDE 19

Time Pixel

17

Adaptive Sampling [Mitchell 91]

slide-20
SLIDE 20

Time Pixel

18

Adaptive Sampling [Mitchell 91]

slide-21
SLIDE 21

Time Pixel

19

Adaptive Sampling [Mitchell 91]

slide-22
SLIDE 22

Time Pixel

20

Adaptive Sampling [Mitchell 91]

slide-23
SLIDE 23

Time Pixel

21

Adaptive Sampling [Mitchell 91]

slide-24
SLIDE 24

Time Pixel

22

Adaptive Sampling [Mitchell 91]

slide-25
SLIDE 25

Time Pixel

23

Adaptive Sampling [Mitchell 91]

slide-26
SLIDE 26

Time Pixel

24

Adaptive Sampling [Mitchell 91]

slide-27
SLIDE 27

Time Pixel

25

Adaptive Sampling [Mitchell 91]

slide-28
SLIDE 28

Time Pixel

26

Adaptive Sampling [Mitchell 91]

slide-29
SLIDE 29

Time Pixel

27

Adaptive Sampling [Mitchell 91]

slide-30
SLIDE 30

Time Pixel

28

Adaptive Sampling [Mitchell 91]

slide-31
SLIDE 31

Time Pixel

29

Adaptive Sampling [Mitchell 91]

slide-32
SLIDE 32

Time Pixel

30

Adaptive Sampling [Mitchell 91]

slide-33
SLIDE 33

Time Pixel

31

Adaptive Sampling [Mitchell 91]

slide-34
SLIDE 34

Time Pixel

32

Adaptive Sampling [Mitchell 91]

slide-35
SLIDE 35

Time Pixel

33

Adaptive Sampling [Mitchell 91]

slide-36
SLIDE 36

Time Pixel

34

Adaptive Sampling [Mitchell 91]

slide-37
SLIDE 37

Time Pixel

35

Adaptive Sampling [Mitchell 91]

slide-38
SLIDE 38

36

Adaptive Sampling [Mitchell 91]

Exact

Sampled

Time Pixel

slide-39
SLIDE 39

Motion Blurred Sphere

37

Samples / Pixel 1 4 16 64 Reference

slide-40
SLIDE 40

Motion Blurred Sphere

38

Samples / Pixel 1 4 16 64 Reference

slide-41
SLIDE 41

39

Adaptive Sampling [Mitchell 91]

Exact

Sampled

Time Pixel

slide-42
SLIDE 42

Key Idea of Our Work

  • Conventional adaptive sampling
  • Only adaptive on pixel axis
  • Multidimensional adaptive sampling
  • Adaptive on all axes

40

slide-43
SLIDE 43

Time Pixel

41

Multidimensional Adaptive Sampling

slide-44
SLIDE 44

Multidimensional Adaptive Sampling

42

Time Pixel

slide-45
SLIDE 45

Multidimensional Adaptive Sampling

43

Time Pixel

slide-46
SLIDE 46

Multidimensional Adaptive Sampling

44

Time Pixel

slide-47
SLIDE 47

Multidimensional Adaptive Sampling

45

Time Pixel

slide-48
SLIDE 48

Multidimensional Adaptive Sampling

46

Time Pixel

slide-49
SLIDE 49

Multidimensional Adaptive Sampling

47

Time Pixel

slide-50
SLIDE 50

Multidimensional Adaptive Sampling

48

Time Pixel

slide-51
SLIDE 51

Multidimensional Adaptive Sampling

49

Time Pixel

slide-52
SLIDE 52

Multidimensional Adaptive Sampling

50

Time Pixel

slide-53
SLIDE 53

Multidimensional Adaptive Sampling

51

Time Pixel

slide-54
SLIDE 54

Multidimensional Adaptive Sampling

52

Time Pixel

slide-55
SLIDE 55

Multidimensional Adaptive Sampling

53

Time Pixel

slide-56
SLIDE 56

Multidimensional Adaptive Sampling

54

Time Pixel

slide-57
SLIDE 57

Multidimensional Adaptive Sampling

55

Time Pixel

slide-58
SLIDE 58

Multidimensional Adaptive Sampling

56

Time Pixel

slide-59
SLIDE 59

Multidimensional Adaptive Sampling

57

Time Pixel

slide-60
SLIDE 60

Multidimensional Adaptive Sampling

58

Time Pixel

slide-61
SLIDE 61

Multidimensional Adaptive Sampling

59

Time Pixel

slide-62
SLIDE 62

Multidimensional Adaptive Sampling

60

Time Pixel

slide-63
SLIDE 63

Multidimensional Adaptive Sampling

61

Time Pixel

slide-64
SLIDE 64

Multidimensional Adaptive Sampling

62

Exact

Sampled

Time Pixel

slide-65
SLIDE 65

Motion Blurred Sphere - Comparison

63

1 4 16 64 Reference [Mitchell 91] Our method

slide-66
SLIDE 66

Motion Blurred Sphere - Comparison

64

1 4 16 64 Reference [Mitchell 91] Our method

slide-67
SLIDE 67

Motion Blurred Sphere

65

1 4 16 64 256 5 10 15 20

Our Method Low Discrepancy Mitchell

slide-68
SLIDE 68
  • Each pixel value, L(x, y), is defined as

multidimensional integration

Rendering is Integration

L (x, y) =

  • · · ·
  • f (x, y, u1, . . . , un) du1 . . . dun

66

slide-69
SLIDE 69

Rendering is Integration

  • Motion blur is 1D integration over time (t)

L (x, y) =

  • f (x, y, t) dt

67

slide-70
SLIDE 70

Rendering is Integration

  • DOF is 2D integration over lens (u, v)

L (x, y) =

  • f (x, y, u, v) dudv

68

slide-71
SLIDE 71

Analytical Solution

69

Time Pixel

L (x) =

  • f (x, t)dt

L (x) x t

slide-72
SLIDE 72

Analytical Solution

70

Time Pixel

L (x) x t L (x) ≈

  • i

wif (x, ti)

slide-73
SLIDE 73

71

Time Pixel

Conventional Adaptive Sampling

L (x) x t L (x) ≈

  • i

wif (x, ti)

slide-74
SLIDE 74

72

Time Pixel

Multidimensional Adaptive Sampling

L (x) x t L (x) ≈

  • i

wif (x, ti)

slide-75
SLIDE 75

Key Features

  • Exploits coherency beyond the image plane
  • Adaptive sampling based on deterministic samples
  • Theory applies to various effects

73

slide-76
SLIDE 76

Results

slide-77
SLIDE 77

Motion Blur (3D) - Reference

75

512 samples 27,488 sec

slide-78
SLIDE 78

Motion Blur (3D) - Our Method

76

8 samples 672.2 sec

slide-79
SLIDE 79

Motion Blur (3D) - [Mitchell 91]

77

12.67 samples 676.4 sec

slide-80
SLIDE 80

Motion Blur (3D) - Comparison

78

Our method 8 samples / pixel 672.2 sec RMS: 0.0034 [Mitchell 91] 12.67 samples / pixel 676.4 sec RMS: 0.0099

slide-81
SLIDE 81

Depth of Field (4D) - Reference‏

79

512 samples 11,960 sec

slide-82
SLIDE 82

Depth of Field (4D) - Our Method

80

16 samples 993 sec

slide-83
SLIDE 83

Depth of Field (4D) - [Mitchell 91]‏

81

38.25 samples 980 sec

slide-84
SLIDE 84

82

Our method 16 samples / pixel 993 sec RMS: 0.132 [Mitchell 91] 38.25 samples / pixel 980 sec RMS: 0.192

Depth of Field (4D) - Comparison

slide-85
SLIDE 85

Sampling Density

83

slide-86
SLIDE 86

Area Light Source (4D)

84

Our Method Mitchell 8 64

Our method Mitchell Samples / Pixel 1 64 Scene Setup

Occluders

Area Light Source

slide-87
SLIDE 87

Area Light Source (4D)

85

Our Method Mitchell 8 64

Our method Mitchell Samples / Pixel 1 64

Low Discrepancy Metropolis

Metropolis

slide-88
SLIDE 88

Area Light Source (4D)

86

Our method Mitchell Metropolis

Occluders

Area Light Source

Samples / Pixel 64

slide-89
SLIDE 89

Motion Blur + DOF (5D)

87

slide-90
SLIDE 90

Future Work

  • More higher dimensional adaptive sampling
  • New adaptive sampling criterion
  • Application to more complex light transport

88

slide-91
SLIDE 91

Conclusion

  • Produces images with less noise by fewer samples
  • Adaptive sampling in multidimensional space
  • Reconstruction from multidimensional samples
  • Applicable to various rendering effects

89

slide-92
SLIDE 92

Acknowledgement

  • NSF CPA 0701992
  • UCSD FW Grid Project
  • NSF Research Infrastructure Grant EIA-0303622
  • All UCSD graphics lab members

90

slide-93
SLIDE 93

Thanks

91