Volume Rendering Multivariate Data to Visualize Meteorological Simulations: A Case Study

1
Volume Rendering Multivariate Data to Visualize
Meteorological Simulations A Case Study
2

• Joe Kniss Charles Hansen
• Scientific Computing and Imaging
• University of Utah
• jmk,hansen_at_cs.utah.edu
• Michel Grenier Tom Robinson
• Canada Meteorological Centre
• Meteorological Service of Canada
• Michel.Grenier, Tom.Robinson_at_ec.gc.ca

3
Outline

• What is Volume Rendering?
• Motivation Background
• Multi-dimensional Transfer Functions
• Interaction Tools
• Shading Hardware
• Parallel Hardware Volume Rendering

4
Overview

• Numerical weather simulation
• Guided by measured data
• Frontal analysis needed
• Multiple data values describe features
• Scalar volume rendering is inadequate
• Current tools are difficult to use

5
Fronts
Overview

• Strong thermal gradient
• Warm/ cold air mass interfaces
• Weather happens here

Frontal movement
warm moist
cold
cold
6
Fronts
Overview

• Warm fronts
• Warm air pushes cold air
• Mild weather

Frontal movement
warm moist
cold
7
Fronts
Overview

• Cold front
• Cold air pushes warm air
• Severe weather

Frontal movement
warm moist
cold
8
Fronts
Overview

• Cold fronts
• Strong convection
• Lightning, hail, tornados

Frontal movement
warm moist
cold
9
Simulation
Overview

• Focused on North America

10
Simulation
Overview

• Focused on North America

11
Simulation
Overview

• Focused on North America
• Variables
• Temperature

12
Simulation
Overview

• Focused on North America
• Variables
• Temperature
• Humidity

13
Simulation
Overview

• Focused on North America
• Variables
• Temperature
• Humidity
• Pressure

14
Simulation
Overview

• Focused on North America
• Variables
• Temperature
• Humidity
• Pressure
• Dew point

15
Simulation
Overview

• Focused on North America
• Variables
• Temperature
• Humidity
• Pressure
• Dew point
• Updated with measured data

16
Goals

• Visualize entire simulation domain
• Identify features using multiple values
• Add derivative measures
• Simple interface
• Realistic lighting

17
Goals

• Visualize entire simulation domain
• Identify features using multiple values
• Add derivative measures
• Simple interface
• Realistic lighting
• Volume rendering is a natural choice
• Multi-dimensional transfer functions
• Dual-domain interaction
• Shadows

18
Previous Methods

• Slice overlays

Temperature
Humidity
19
Previous Methods

• Slice overlays

Temperature
Overlay
Humidity
20
Previous Methods

• Slice overlays

Temperature
Humidity
Analysis
21
Previous Methods

• Slice overlays

Temperature
Humidity
Analysis
Enhancements
22
Previous Methods

• Slice overlays

23
Previous Methods

• Transfer function sharing
• Set each optical property using different data

Ctemp Ahumidity
Cpressure Ahumidity
Vis5D- Bill Hibbard, SSEC
24
Previous Methods

• Multi-dimensional transfer functions
• Scalar data and derivatives

Display of Surfaces, Levoy 1988
25
Previous Methods

• Multi-dimensional transfer functions
• Scalar data and derivatives

Display of Surfaces, Levoy 1988
Visualization 2001
26
Previous Methods

• Multi-dimensional transfer functions
• Multivariate data (MRI)

T2
Laidlaw 1995
T1
27
Multivariate data

• Simulation variables

28
Multivariate data

• Simulation variables

Pressure
Humidity
Temperature
Humidity
Wet bulb temp
Pressure
Temperature
Temperature
29
Classification

• 1D verses 2D transfer functions

1D transfer function
2D transfer function
30
Classification

• Exploration Transfer function domain

31
Classification

• Exploration Transfer function domain

32
Classification

• Exploration Transfer function domain

33
Classification

• Exploration Transfer function domain

34
Classification

• Exploration Transfer function domain

35
Classification

• Exploration Transfer function domain

36
Classification

• Exploration Transfer function domain

37
Classification

• Exploration Transfer function domain

Unintuitive
Tedious
38
Classification

• Exploration Probing

39
Classification

• Exploration Probing

40
Classification

• Exploration Probing

41
Classification

• Exploration Probing

42
Classification

• Exploration Probing

43
Classification

• Exploration Dual-domain interaction

44
Classification

• Exploration Dual-domain interaction

45
Classification

• Exploration Dual-domain interaction

46
Classification

• Exploration Dual-domain interaction

47
Classification

• Exploration Dual-domain interaction

48
Classification

• Exploration Dual-domain interaction

49
Classification

• Exploration Dual-domain interaction

50
Classification

• Exploration Dual-domain interaction

51
Classification

• Exploration Dual-domain interaction

52
Classification

• Strong gradient at frontal boundaries

Frontal movement
warm moist
cold
cold
53
Classification

• Strong gradient at frontal boundaries
• A derivative measure can help

Frontal movement
warm moist
cold
cold
54
Multivariate data

• Simulation variables derivative measures

Humidity
Humidity
Temperature
Gradient magnitude
Humidity
Gradient magnitude
Gradient magnitude
Temperature
Temperature
55
Multivariate derivatives

• Sum individual gradients and magnitudes
• Simple implementation

2 valued field gradients
56
Multivariate derivatives

• Sum individual gradients and magnitudes
• Simple implementation

2 valued field gradients
Multi-gradient
57
Multivariate derivatives

• Sum individual gradients and magnitudes
• Simple implementation
• Not robust for normal generation

2 valued field gradients
58
Multivariate derivatives

• Sum individual gradients and magnitudes
• Simple implementation
• Not robust for normal generation

2 valued field gradients
Multi-gradient
59
Multivariate derivatives

• Sum individual gradients and magnitudes
• Simple implementation
• Not robust for normal generation

2 valued field gradients
Multi-gradient
60
Multivariate derivatives

• Sum individual gradients and magnitudes
• Simple implementation
• Not robust for normal generation
• Worse for more variables

5 valued field gradients
61
Multivariate derivatives

• Sum individual gradients and magnitudes
• Simple implementation
• Not robust for normal generation
• Worse for more variables

?
5 valued field gradients
Multi-gradient
62
Multivariate derivatives

• Local contrast
• Scalar data

Volume space
Scalar data space
63
Multivariate derivatives

• Local contrast
• Scalar data

Volume space
Scalar data space
64
Multivariate derivatives

• Local contrast
• Scalar data

Volume space
Scalar data space
65
Multivariate derivatives

• Local contrast
• Scalar data

Volume space
Scalar data space
66
Multivariate derivatives

• Local contrast
• Scalar data

Volume space
Scalar data space
67
Multivariate derivatives

• Local contrast
• Scalar data

68
Multivariate derivatives

• Local contrast
• Scalar data

Scalar value
Scalar value
69
Multivariate derivatives

• Local contrast
• Scalar data
• Analyze

Scalar value
Scalar value
70
Multivariate derivatives

• Local contrast
• Multivariate data

Volume space
Multivariate data space
71
Multivariate derivatives

• Local contrast
• Multivariate data

Volume space
Multivariate data space
72
Multivariate derivatives

• Local contrast
• Multivariate data

Volume space
Multivariate data space
73
Multivariate derivatives

• Local contrast
• Multivariate data

Volume space
Multivariate data space
74
Multivariate derivatives

• Local contrast
• Multivariate data

3xN matrix
75
Multivariate derivatives

• Local contrast
• Multivariate data

76
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro

77
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro

3x3 matrix
78
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani

79
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani
• Analyze G

80
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani
• Analyze G, total change

81
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani
• Analyze G, total change

82
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani
• Analyze G, total change, eigenvectors

83
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani
• Analyze G, total change, eigenvectors
• L2 norm

84
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani
• Analyze G, total change, eigenvectors
• L2 norm
• Scale each variables derivatives

85
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani
• Analyze G, total change, eigenvectors
• L2 norm
• Scale each variables derivatives
• Same as gradient for single variable functions

86
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani
• Analyze G, total change, eigenvectors
• L2 norm
• Scale each variables derivatives
• Same as gradient for single variable functions
• Eigenvectors only provide orientation

87
Multivariate derivatives

• Local contrast
• Multivariate data
• Squared norm DiZenzo, Cumani, Sapiro
• Squared Local Contrast Cumani
• Analyze G, total change, eigenvectors
• L2 norm
• Scale each variables derivatives
• Same as gradient for single variable functions
• Eigenvectors only provide orientation
• Use shadows instead of surface shading

88
Classification

• Using multivariate data and derivatives
• Default transfer function

Lower threshold for gradient magnitude
Upper threshold for gradient magnitude
gradient magnitude
89
Classification

• Using multivariate data and derivatives
• Default transfer function

Lower threshold for gradient magnitude
Upper threshold for gradient magnitude
gradient magnitude
90
Classification

• Using multivariate data and derivatives
• Default transfer function

Lower threshold for gradient magnitude
Upper threshold for gradient magnitude
gradient magnitude
91
Classification

• Using multivariate data and derivatives
• Default transfer function
• Better specificity

No gradient enhancement
92
Classification

• Using multivariate data and derivatives
• Default transfer function
• Better specificity

No gradient enhancement
With gradient enhancement
93
Classification

• Using multivariate data and derivatives
• Default transfer function
• Better specificity

No gradient enhancement
With gradient enhancement
94
Results

• Frontal analysis is subjective

95
Results

• Frontal analysis is subjective
• Volume rendering provides objective guidance

96
Results

• Frontal analysis is subjective
• Volume rendering provides objective guidance

97
Results

• Frontal analysis is subjective
• Volume rendering provides objective guidance

98
Results

• Frontal analysis is subjective
• Volume rendering provides objective guidance
• Multi-dimensional transfer functions help
• Greater specificity
• Greater flexibility
• New kind of analysis

99
Results

• Frontal analysis is subjective
• Volume rendering provides objective guidance
• Multi-dimensional transfer functions help
• Greater specificity
• Greater flexibility
• New kind of analysis
• Meteorologists prefer top view

100
Results

• Video

101
Results

• Other application areas

Medical Imaging
Color Data
102
Results

• Multiple MRI pulse sequences

T1
T2
Utah HLRS, Stuttgart
103
Results

• RGB color data (cryosection)

Blue
Red
Green
TVCG July-Sep 2002
104
Future Work

• Higher dimensional transfer functions
• Implicit classification functions
• Dual-domain interaction is important
• Time varying transfer functions

105
Future Work

• Higher dimensional transfer functions
• Implicit classification functions
• Dual-domain interaction is important
• Time varying transfer functions
• Vector volume rendering glyphs

106
Future Work

• Higher dimensional transfer functions
• Implicit classification functions
• Dual-domain interaction is important
• Time varying transfer functions
• Vector volume rendering glyphs
• Terrain

107
Future Work

• Higher dimensional transfer functions
• Implicit classification functions
• Dual-domain interaction is important
• Time varying transfer functions
• Vector volume rendering glyphs
• Terrain
• 3D illustration and labeling tools

108
Future Work

• Higher dimensional transfer functions
• Implicit classification functions
• Dual-domain interaction is important
• Time varying transfer functions
• Vector volume rendering glyphs
• Terrain
• 3D illustration and labeling tools
• Better default transfer functions
• Automation animation

109
Future Work

• Realistic clouds, with David Ebert

110
Shadows
Hardware
Sample ri (s)
l0
Attenuate light through the volume
111
Incremental Shadows
Eye
Half angle slicing good from either point of
view
112
Incremental Shadows
Similar aspect ratio from both points of view
113
Incremental Shadows

Slice pass 1
114
Incremental Shadows
Slice pass 2
115
Incremental Shadows

• Advantages
• Screen space shadows
• No leakage
• Use render to texture to optimize
• Shades perturbed volumes
• Simple implementation
• Disadvantages
• Aliasing at sharp opacity changes
• Fix with slightly larger light buffer

116
Shadows
Hardware

• Shadows and Surface shading
• Modulate surface shading by gradient magnitude
• Add surface scalar to the TF

Homogeneous w/ shading
Homogeneous w/out shading
Shadows surface shading
117
Background

• 5 stages
• Preprocessing filtering, resizing, bricking,
  derivatives
• Reconstruction interpolation
• Classification data value -gt optical properties
• Shading surface shading volume shading
• Compositing numerical integration, blending
  operations

Pre-classification
Post-classification
118
Transfer Function
Hardware

• Separability

General 2D

X
1D
1D
2D
119
Transfer Function
Hardware

• Sepparability

General 2D

X
1D
1D
2D
120
Shading
Hardware

• Interpolation de-normalization
• Sqrt() not yet available
• Solution Cube Map dependent texture
• Unit normals not required
• Arbitrary lighting
• Slower than dot-product

Cube map
See DirectX 9 spec
121
Shading
Hardware

• Surface shading is not enough
• Poor use of valuable texture memory
• Cannot light homogenious regions
• Cannot light multi-valued volumes
• Solution
• Faux shading
• Shadows
• Smart surface shading

122
Faux Shading
Hardware

• Modify color based on opacity
• Best for discretely classified regions

opacity
opacity
color
color
Standard
Faux
123
Other Axes
Transfer Function

• Derivatives

Derivative of a multi-field
3x3 Matrix!
Squared Norm
Di Zenzo, Cumani, Sapiro
124
Fronts

• Strong thermal gradient

Frontal movement
warm moist
cold
cold
125
Motivation

• Most information through eyes
• Visualization is key
• Volume rendering is flexible
• Medical Imaging
• Simulation Data
• Synthetic Data

126
Multivariate derivatives

• Local contrast
• Multivariate data