SURE-based Optimization for Adaptive Sampling and Reconstruction - - PowerPoint PPT Presentation

▶
sure based optimization for adaptive sampling and
SMART_READER_LITE
LIVE PREVIEW

SURE-based Optimization for Adaptive Sampling and Reconstruction - - PowerPoint PPT Presentation

Jan 04, 2024 313 likes •1.13k views

SURE-based Optimization for Adaptive Sampling and Reconstruction Supplementary Materials Tzu-Mao Li Yu-Ting Wu Yung-Yu Chuang National Taiwan University PART I Equal-Time Comparison Compared Methods: Monte Carlo Greedy Error

slide-1
SLIDE 1

SURE-based Optimization for Adaptive Sampling and Reconstruction

Supplementary Materials

Tzu-Mao Li Yu-Ting Wu Yung-Yu Chuang National Taiwan University

slide-2
SLIDE 2

PART I Equal-Time Comparison

Compared Methods:

  • Monte Carlo
  • Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011]
  • Random Parameter Filtering [Sen and Darabi, ACMTOG 2012]
  • SURE-based Optimization (our approach, using cross bilateral filters)
slide-3
SLIDE 3

SPONZA

Global Illumination (Path Tracing) Motion Blur

1600 x 1200

slide-4
SLIDE 4

SPONZA Equal-time Monte Carlo, 68 spp, 890.5 sec.

slide-5
SLIDE 5

SPONZA Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 63.84 spp, 906.2 sec.

slide-6
SLIDE 6

SPONZA Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 16 spp, 1676.1 sec.

slide-7
SLIDE 7

SPONZA SURE-based Optimization (Our Approach), 63.24 spp, 896.0 sec.

slide-8
SLIDE 8

SPONZA Reference, 8192 spp

slide-9
SLIDE 9

TOWN

Environment Lighting Area Lighting Motion Blur

800 x 600

slide-10
SLIDE 10

TOWN Equal-time Monte Carlo, 82 spp, 59.9 sec.

slide-11
SLIDE 11

TOWN Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 51.82 spp, 61.8 sec.

slide-12
SLIDE 12

TOWN Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 8 spp, 272.4 sec.

slide-13
SLIDE 13

TOWN SURE-based Optimization (Our Approach), 39.79 spp, 59.6 sec.

slide-14
SLIDE 14

TOWN Reference, 4096 spp

slide-15
SLIDE 15

SIBENIK

Global Illumination (One-Bounce Path Tracing) Depth of Field

1024 x 1024

slide-16
SLIDE 16

SIBENIK Equal-time Monte Carlo, 44 spp, 140.0 sec.

slide-17
SLIDE 17

SIBENIK Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 39.86 spp, 135.0 sec.

slide-18
SLIDE 18

SIBENIK Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 8 spp, 363.0 sec.

slide-19
SLIDE 19

SIBENIK SURE-based Optimization (Our Approach), 26.69 spp, 140 sec.

slide-20
SLIDE 20

SIBENIK Reference, 4096 spp

slide-21
SLIDE 21

TEAPOT

Environment Lighting Glossy Reflection

800 x 800

slide-22
SLIDE 22

TEAPOT Equal-time Monte Carlo, 35 spp, 42.0 sec.

slide-23
SLIDE 23

TEAPOT Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 23.96 spp, 44.3 sec.

slide-24
SLIDE 24

TEAPOT Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 8 spp, 374.4 sec.

slide-25
SLIDE 25

TEAPOT SURE-based Optimization (Our Approach), 8 spp, 40.4 sec.

slide-26
SLIDE 26

TEAPOT Reference, 4096 spp

slide-27
SLIDE 27

GARGOYLE

Global Illumination (One-Bounce Path Tracing)

1024 x 1024

slide-28
SLIDE 28

GARGOYLE Equal-time Monte Carlo, 56 spp, 161.7 sec.

slide-29
SLIDE 29

GARGOYLE Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 43.92 spp, 167.4 sec.

slide-30
SLIDE 30

GARGOYLE Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 8 spp, 608.3 sec.

slide-31
SLIDE 31

GARGOYLE SURE-based Optimization (Our Approach), 30.90 spp, 160.0 sec.

slide-32
SLIDE 32

GARGOYLE Reference, 4096 spp

slide-33
SLIDE 33

SANMIGUEL

Global Illumination (Path Tracing)

1580 x 986

slide-34
SLIDE 34

SANMIGUEL Equal-time Monte Carlo, 70 spp, 1209.4 sec.

slide-35
SLIDE 35

SANMIGUEL Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 63.59 spp, 1239.9 sec.

slide-36
SLIDE 36

SANMIGUEL Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 16 spp, 2617.9 sec.

slide-37
SLIDE 37

SANMIGUEL SURE-based Optimization (Our Approach), 61.69 spp, 1228.9 sec.

slide-38
SLIDE 38

SANMIGUEL Reference, 8192 spp

slide-39
SLIDE 39

PART II Equal-Sample Comparison

Compared Methods:

  • Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011]
  • Random Parameter Filtering [Sen and Darabi, ACMTOG 2012]
  • SURE-based Optimization (our approach, using cross bilateral filters)
slide-40
SLIDE 40

SPONZA

Global Illumination (Path Tracing) Motion Blur

1600 x 1200

slide-41
SLIDE 41

SPONZA Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 16 spp, 210.0 sec.

slide-42
SLIDE 42

SPONZA Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 16 spp, 1676.1 sec.

slide-43
SLIDE 43

SPONZA SURE-based Optimization (Our Approach), 16 spp, 273.3 sec.

slide-44
SLIDE 44

SPONZA Reference, 8192 spp

slide-45
SLIDE 45

TOWN

Environment Lighting Area Lighting Motion Blur

800 x 600

slide-46
SLIDE 46

TOWN Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 8 spp, 9.4 sec.

slide-47
SLIDE 47

TOWN Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 8 spp, 272.4 sec.

slide-48
SLIDE 48

TOWN SURE-based Optimization (Our Approach), 8 spp, 20.0 sec.

slide-49
SLIDE 49

TOWN Reference, 4096 spp

slide-50
SLIDE 50

SIBENIK

Global Illumination (One-Bounce Path Tracing) Depth of Field

1024 x 1024

slide-51
SLIDE 51

SIBENIK Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 8 spp, 27.6 sec.

slide-52
SLIDE 52

SIBENIK Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 8 spp, 363.0 sec.

slide-53
SLIDE 53

SIBENIK SURE-based Optimization (Our Approach), 8 spp, 64.2 sec.

slide-54
SLIDE 54

SIBENIK Reference, 4096 spp

slide-55
SLIDE 55

TEAPOT

Environment Lighting Glossy Reflection

800 x 800

slide-56
SLIDE 56

TEAPOT Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 8 spp, 14.1 sec.

slide-57
SLIDE 57

TEAPOT Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 8 spp, 374.4 sec.

slide-58
SLIDE 58

TEAPOT SURE-based Optimization (Our Approach), 8 spp, 40.4 sec.

slide-59
SLIDE 59

TEAPOT Reference, 4096 spp

slide-60
SLIDE 60

GARGOYLE

Global Illumination (One-Bounce Path Tracing)

1024 x 1024

slide-61
SLIDE 61

GARGOYLE Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 8 spp, 28.6 sec.

slide-62
SLIDE 62

GARGOYLE Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 8 spp, 608.3 sec.

slide-63
SLIDE 63

GARGOYLE SURE-based Optimization (Our Approach), 8 spp, 68.3 sec.

slide-64
SLIDE 64

GARGOYLE Reference, 4096 spp

slide-65
SLIDE 65

SANMIGUEL

Global Illumination (Path Tracing)

1580 x 986

slide-66
SLIDE 66

SANMIGUEL Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011], 16 spp, 304.4 sec.

slide-67
SLIDE 67

SANMIGUEL Random Parameter Filtering [Sen and Darabi, ACMTOG 2012], 16 spp, 2617.9 sec.

slide-68
SLIDE 68

SANMIGUEL SURE-based Optimization (Our Approach), 16 spp, 336.3 sec.

slide-69
SLIDE 69

SANMIGUEL Reference, 8192 spp

slide-70
SLIDE 70

PART III Equal-Time Comparison for Isotropic Gaussian Filters

Compared Methods:

  • Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011]
  • SURE-based Optimization (our approach, using isotropic Gaussian filters)
slide-71
SLIDE 71

TOASTERS

Area Lighting Depth of Field

1024 x 1024

slide-72
SLIDE 72

TOASTERS Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011]

slide-73
SLIDE 73

TOASTERS SURE-based Optimization (Our Approach), using Isotropic Gaussian Filters

slide-74
SLIDE 74

TOASTERS Reference, 4096 spp

slide-75
SLIDE 75

TOASTERS – Scale Selection Map Greedy Error Minimization [Rousselle et al., SIGGRAPH Asia 2011]

slide-76
SLIDE 76

TOASTERS - Scale Selection Map SURE-based Optimization (Our Approach), using Isotropic Gaussian Filters

slide-77
SLIDE 77

PART IV Equal-Time Comparison for Cross Non-local Means Filters

Compared Methods:

  • Global cross non-local means filters
  • SURE-based Optimization (our approach, using cross non-local means filters)
slide-78
SLIDE 78

TOWN

Environment Lighting Area Lighting Motion Blur

800 x 600

slide-79
SLIDE 79

TOWN Global Non-local Means Filter, 41.2 spp

slide-80
SLIDE 80

TOWN SURE-based Optimization (Our Approach), using Cross Non-local Means Filters, 41.2 spp, 244.7 sec.

slide-81
SLIDE 81

TOWN Reference, 4096 spp