Visualization Knowledge (VisKo): Leveraging the Semantic Web to Support Visualization

University of Texas at El PasoComputer Science

Velocity Model Visualization

• A set of isosurfaces– extracted from a seismic velocity model– Covers a region in southern New Mexico

8 km/s

3 km/s

Dept

h

Visualizing the Velocity Model

• Generated by custom Java application– relied on Visualization Toolkit (VTK) for rendering– VTK was developed for rendering 3D visualizations– VTK is supported by Sandia, Los Alamos, ARL, and others

• Writing a custom visualization application:– may rely on third party package to support rendering– may need to perform some transformations on input

dataset before rendering

Program For Velocity Visualization

vtkImageReader rdr = new vtkImageReader();rdr.SetFileName(inputDatasetFilePath);rdr.SetDataScalarTypeToUnsignedShort();rdr.SetDataByteOrderToLittleEndian();rdr.SetFileDimensionality(3);rdr.SetDataOrigin(0,0,0);rdr.SetDataSpacing(1,1,1);rdr.SetDataExtent(0,230,0,25,0,68);rdr.SetNumberOfScalarComponents(1);rdr.FileLowerLeftOn();rdr.Update();

vtkContourFilter contours = new vtkContourFilter();contours.SetInput(rdr.GetOutput());contours.GenerateValues(35,0.0,9000.0);

vtkPolyDataMapper contMapper = new vtkPolyDataMapper();contMapper.SetInput(contours.GetOutput());contMapper.SetScalarRange(0.0,9000.0);

vtkActor contActor = new vtkActor();contActor.SetMapper(contMapper);contActor.RotateX(105);

vtkRenderer ren1 = new vtkRenderer();ren1.AddActor(contActor);ren1.AddActor2D(outlineActor);ren1.SetBackground(1,1,1);

vtkRenderWindow renWin = new vtkRenderWindow();renWin.SetOffScreenRendering(1);renWin.AddRenderer(ren1);renWin.SetSize(300,300);renWin.Render();

vtkJPEGWriter img = new vtkJPEGWriter();img.SetInputConnection(renWin.GetOutputPort());img.SetFileName(outputDatasetFilePath);img.SetQuality(100);

What Information is Needed?

• Regarding visualization, users need to know:– What view suits their data (e.g., contours, map, surfaces)– What properties should the view exhibit (e.g., orientation,

projection, color, size)

• Regarding datasets, users need to know:– The semantic type of the data (e.g., gravity, velocity)– The type of data helps with setting parameter arguments– The format the data is encoded in (e.g., netCDF, ESRI)

What Information is Needed?

• Regarding third party rendering software:– Can it generate the required renderings?– Can it ingest my data in the format it resides in?– Does it satisfy my performance requirements?– What language can I use to interface with it?– What dependent packages do I need to install?

Where to Get Information

• Regarding the view:– some user may prefer to see isosurfaces whereas another

user prefers volume renderings (its personal)• Regarding the data:– provenance may declare what format and type of dataset– Provenance Markup Language (PML) has fields for

recording both type and format• Regarding the third party rendering software:– Reading, reading, and more reading– May have a visualization expert assist you

Writing the Program

• Once you have answers for the previous questions, you can begin coding

• Some portion of the code will transform input datasets into formats that can be rendered– Most renderers format specific (exception Protovis)

• The rest of the code will:– Gather suitable arguments for the rendering– Invoking the renderer with the arguments

Overview of the Visualization Program

• The visualization program that generated the velocity model visualization:– Relies on VTK for rendering– Renders the data as isosurfaces– Ingests data in format binaryFloatArray– transforms binaryFloatArray to VTKImageData– Ingests data of type 3DVelocityModel– Is a serial application– Is written in Java

Presentation Overview

1. Toolkits and Visualization Pipelines2. Visualization Query3. Automated Generation of Visualization pipelines4. Ontological Description of Visualization Pipelines5. Execution of Visualization Pipelines6. Giovanni and VisKo

Visualization Toolkits

• Visualization Toolkit (VTK) was used to render the velocity visualization

• VTK is a toolkit that provides functions such as:– Filtering– Gridding/interpolating– Mapping (i.e., transform data into views like isosurfaces)– Rendering the views

• Functions are referred to as operators– Generic mapping tools (GMT): 60 operators– VTK: hundreds of operators

Visualization Pipeline Model• VTK requires that users write pipelines– the output of an operator feeds into the operator next in

the pipeline sequence– first operator in pipeline is usually data reader– final operator in pipeline is usually renderer

• Thus the Java program that visualizes the velocity model can be seen as a pipeline of VTK operators

• It is up to the users to write these pipelines…

VTK Java Pipeline For Velocity Model

vtkImageReader rdr = new vtkImageReader();rdr.SetFileName(inputDatasetFilePath);rdr.SetDataScalarTypeToUnsignedShort();rdr.SetDataByteOrderToLittleEndian();rdr.SetFileDimensionality(3);rdr.SetDataOrigin(0,0,0);rdr.SetDataSpacing(1,1,1);rdr.SetDataExtent(0,230,0,25,0,68);rdr.SetNumberOfScalarComponents(1);rdr.FileLowerLeftOn();rdr.Update();

vtkContourFilter contours = new vtkContourFilter();contours.SetInput(rdr.GetOutput());contours.GenerateValues(35,0.0,9000.0);

vtkPolyDataMapper contMapper = new vtkPolyDataMapper();contMapper.SetInput(contours.GetOutput());contMapper.SetScalarRange(0.0,9000.0);

vtkActor contActor = new vtkActor();contActor.SetMapper(contMapper);contActor.RotateX(105);

vtkRenderer ren1 = new vtkRenderer();ren1.AddActor(contActor);ren1.AddActor2D(outlineActor);ren1.SetBackground(1,1,1);

vtkRenderWindow renWin = new vtkRenderWindow();renWin.SetOffScreenRendering(1);renWin.AddRenderer(ren1);renWin.SetSize(300,300);renWin.Render();

vtkJPEGWriter img = new vtkJPEGWriter();img.SetInputConnection(renWin.GetOutputPort());img.SetFileName(outputDatasetFilePath);img.SetQuality(100);

Pipeline of Visualization Operators

vtkImageReader rdr = new vtkImageReader();rdr.SetFileName(inputDatasetFilePath);rdr.SetDataScalarTypeToUnsignedShort();rdr.SetDataByteOrderToLittleEndian();rdr.SetFileDimensionality(3);rdr.SetDataOrigin(0,0,0);rdr.SetDataSpacing(1,1,1);rdr.SetDataExtent(0,230,0,25,0,68);rdr.SetNumberOfScalarComponents(1);rdr.FileLowerLeftOn();rdr.Update();

vtkContourFilter contours = new vtkContourFilter();contours.SetInput(rdr.GetOutput());contours.GenerateValues(35,0.0,9000.0);

vtkPolyDataMapper contMapper = new vtkPolyDataMapper();contMapper.SetInput(contours.GetOutput());contMapper.SetScalarRange(0.0,9000.0);

vtkActor contActor = new vtkActor();contActor.SetMapper(contMapper);contActor.RotateX(105);

vtkRenderer ren1 = new vtkRenderer();ren1.AddActor(contActor);ren1.AddActor2D(outlineActor);ren1.SetBackground(1,1,1);

vtkRenderWindow renWin = new vtkRenderWindow();renWin.SetOffScreenRendering(1);renWin.AddRenderer(ren1);renWin.SetSize(300,300);renWin.Render();

vtkJPEGWriter img = new vtkJPEGWriter();img.SetInputConnection(renWin.GetOutputPort());img.SetFileName(outputDatasetFilePath);img.SetQuality(100);

Op 1

Op 2

Op 3

Op 5

Op 6

Op 7

Op 8

Different Types of Operators

vtkImageReader rdr = new vtkImageReader();rdr.SetFileName(inputDatasetFilePath);rdr.SetDataScalarTypeToUnsignedShort();rdr.SetDataByteOrderToLittleEndian();rdr.SetFileDimensionality(3);rdr.SetDataOrigin(0,0,0);rdr.SetDataSpacing(1,1,1);rdr.SetDataExtent(0,230,0,25,0,68);rdr.SetNumberOfScalarComponents(1);rdr.FileLowerLeftOn();rdr.Update();

vtkContourFilter contours = new vtkContourFilter();contours.SetInput(rdr.GetOutput());contours.GenerateValues(35,0.0,9000.0);

vtkPolyDataMapper contMapper = new vtkPolyDataMapper();contMapper.SetInput(contours.GetOutput());contMapper.SetScalarRange(0.0,9000.0);

vtkActor contActor = new vtkActor();contActor.SetMapper(contMapper);contActor.RotateX(105);

vtkRenderer ren1 = new vtkRenderer();ren1.AddActor(contActor);ren1.AddActor2D(outlineActor);ren1.SetBackground(1,1,1);

vtkRenderWindow renWin = new vtkRenderWindow();renWin.SetOffScreenRendering(1);renWin.AddRenderer(ren1);renWin.SetSize(300,300);renWin.Render();

vtkJPEGWriter img = new vtkJPEGWriter();img.SetInputConnection(renWin.GetOutputPort());img.SetFileName(outputDatasetFilePath);img.SetQuality(100);

Op 1

Op 2

Op 3

Op 5

Op 6

Op 7

Op 8

Transformer

View Mapper

Renderer

Transformer

Transformer

Transformer

Transformer

Operators are Parameterized

vtkImageReader rdr = new vtkImageReader();rdr.SetFileName(inputDatasetFilePath);rdr.SetDataScalarTypeToUnsignedShort();rdr.SetDataByteOrderToLittleEndian();rdr.SetFileDimensionality(3);rdr.SetDataOrigin(0,0,0);rdr.SetDataSpacing(1,1,1);rdr.SetDataExtent(0,230,0,25,0,68);rdr.SetNumberOfScalarComponents(1);rdr.FileLowerLeftOn();rdr.Update();

vtkContourFilter contours = new vtkContourFilter();contours.SetInput(rdr.GetOutput());contours.GenerateValues(35,0.0,9000.0);

vtkPolyDataMapper contMapper = new vtkPolyDataMapper();contMapper.SetInput(contours.GetOutput());contMapper.SetScalarRange(0.0,9000.0);

vtkActor contActor = new vtkActor();contActor.SetMapper(contMapper);contActor.RotateX(105);

vtkRenderer ren1 = new vtkRenderer();ren1.AddActor(contActor);ren1.AddActor2D(outlineActor);ren1.SetBackground(1,1,1);

vtkRenderWindow renWin = new vtkRenderWindow();renWin.SetOffScreenRendering(1);renWin.AddRenderer(ren1);renWin.SetSize(300,300);renWin.Render();

vtkJPEGWriter img = new vtkJPEGWriter();img.SetInputConnection(renWin.GetOutputPort());img.SetFileName(outputDatasetFilePath);img.SetQuality(100);

Op 1

Op 2

Op 3

Op 5

Op 6

Op 7

Op 8

P1

P2

P5

P6

P7

P8

P9

P3

P4

What Kind of Skills are Currently required by a user?

• Knows about different views• Knows what toolkits support a particular view• Knows what toolkits operate on a particular data• Knows how to install a visualization toolkit• Knows what language the toolkit is built on• Knows what operators need to compose a pipeline• Knows suitable arguments for the operators• Knows how to develop software

What Kind of Skills are Currently required by a user?

• Knows about different views

• Knows what toolkits support a particular view• Knows what toolkits operate on a particular data

• Knows how to install a visualization toolkit• Knows what language the toolkit is built on• Knows what operators need to compose a pipeline• Knows suitable arguments for the operators• Knows how to develop software

scientist

VisualizationExpert

Engineer

Presentation Overview

1. Toolkits and Visualization Pipelines2. Visualization Query3. Automated Generation of Visualization pipelines4. Ontological Description of Visualization Pipelines5. Execution of Visualization Pipelines6. Giovanni and VisKo

Declarative Requests

• Many user skills needed to visualize data• Aside from cognitive aspects of visualization, a large

part of the problem is engineering• Stems from fact that we generate visualizations

imperatively (i.e., write code)

Can we provide a means for users to generate visualizations declaratively (i.e., specify what visualization they

want without having to code)?

Visualization Query

(AND (hasView ?VIS isosurfaces)(hasContent ?VIS http://vel.3d)(hasFormat ?VIS floatArray)(hasType ?VIS velocityData)(viewedBy ?VIS mozilla-firefox)(hasValue numContours 36))

Desired View URL of Data to visualize

Semantic Type of datasetWDO types (UTEP)

Format of dataset

Parameter Argument Viewer

• The velocity model visualization was actually a result of a visualization query

Requested Visualization

Visualization Queries and SQL

• Visualization queries mirror SQL queries– query request is specified declaratively– request is then translated into a query plan– query plan computes the result requested by the query

• Information specified in visualization queries is used to derive pipelines rather than query plans

• The pipeline in turn generates the visualization requested in the query

Visualization Query Challenges

What kind of knowledge is needed to generate pipelines that answer

visualization queries?

What infrastructure can leverage the knowledge to support the generation

and execution of the pipelines?

VisKo Project Claims

• VisKo supplements user skills with visualization knowledge to compose pipelines

• VisKo is a framework for:– Encoding user skills into visualization knowledge– managing visualization knowledge– Executing pipelines

Old Way vs VisKo

vtkImageReader rdr = new vtkImageReader();rdr.SetFileName(inputDatasetFilePath);rdr.SetDataScalarTypeToUnsignedShort();rdr.SetDataByteOrderToLittleEndian();rdr.SetFileDimensionality(3);rdr.SetDataOrigin(0,0,0);rdr.SetDataSpacing(1,1,1);rdr.SetDataExtent(0,230,0,25,0,68);rdr.SetNumberOfScalarComponents(1);rdr.FileLowerLeftOn();rdr.Update();

vtkContourFilter contours = new vtkContourFilter();contours.SetInput(rdr.GetOutput());contours.GenerateValues(35,0.0,9000.0);

vtkPolyDataMapper contMapper = new vtkPolyDataMapper();contMapper.SetInput(contours.GetOutput());contMapper.SetScalarRange(0.0,9000.0);

(AND (hasView ?VIS isosurfaces)(hasContent ?VIS http://vel.3d)(hasFormat ?VIS floatArray)(hasType ?VIS velocityData)(viewedBy ?VIS mozilla-firefox)(hasValue numContours 36))

Presentation Overview

1. Toolkits and Visualization Pipelines2. Visualization Query3. Automated Generation of Visualization pipelines4. Ontological Description of Visualization Pipelines5. Execution of Visualization Pipelines6. Giovanni and VisKo

A Pipeline Synthesized by VisKoda

ta fl

ow

operators

VisKo Pipeline Composition

• The query tells the system:– the input format– the target view

• From target view:– Identify operator that generates view (i.e. view mapper)

• From operator:– identify format it operates on (i.e., target format)

• Find sequence of operators that transforms the input format to target format

Information From Query

hasView

BinaryFloatArray

hasF

orm

at

Dataset

Information from Query

Isosurfaces view

Knowledge about Toolkit Operators

vtkContour

Filter

vtkPolyData

in format

out f

orm

at

generatesViewhasView

vtkImageData

BinaryFloatArray

hasF

orm

at

Dataset

Information from Query Knowledge about toolkit operators

Isosurfaces view

The Knowledge is Linked

vtkContour

Filter

vtkPolyData

in format

out f

orm

at

generatesViewhasView

vtkImageData

BinaryFloatArray

hasF

orm

at

Dataset

Information from Query Knowledge about toolkit operators

Isosurfaces view

Both query and operator reference “isosurfaces”

Format Transformation?

vtkContour

Filter

vtkPolyData

in format

out f

orm

at

generatesViewhasView

vtkImageData

BinaryFloatArray

hasF

orm

at

Dataset

Information from Query Knowledge about toolkit operators

Isosurfaces view

Required Pipeline

vtkContour

Filter

vtkPolyData

in format

out f

orm

at

generatesViewhasView

vtkImageData

BinaryFloatArray

hasF

orm

at

Dataset

Information from Query Knowledge about toolkit operators

Isosurfaces view

Multiple Results Example

Visualization Query

No view specified!

Input format is ESRI Gridded

Data is of type Gravity Data

Multiple Results Example

Query Results

VisKo was able to generate three different pipelines, given the query and visualization

knowledge currently loaded

Multiple Visualizations Example

Query Results

Multiple Visualizations ExampleVisualization Query

Query Results

PDF vs PNG Format

View was left unspecified in query, so system visualized data by any means

Composition by Rules

• Pipeline composition is actually derived through application of rules– rules are applied to statements comprising our

visualization knowledge– rules are simple horn clauses

• Rules are not described in this seminar• Before we can apply rules, we need to know what

statements comprise our knowledge base

Presentation Overview

1. Toolkits and Visualization Pipelines2. Visualization Query3. Automated Generation of Visualization pipelines4. Ontological Description of Visualization Pipelines5. Execution of Visualization Pipelines6. Giovanni and VisKo

VisKo Visualization Language

• VisKo provides a language to describe operators and how they can be composed into pipelines

• The language’s expressivity is focused on describing:– Views and view properties– Different operator types– Parameters associated with operators

• The language is defined by ontologies encoded in Ontology Web Language (OWL)

VisKo Language Layers• The VisKo language encompasses three different

ontologies to describe toolkit operators from different perspectives

Visko-Service(services, parameters, and types)

Visko-Operator(operator function + composition rules)

Visko-Views(views and properties)

Execution

Visualization

Spec

trum

Encoding Velocity Model View

• Velocity model is visualized as a set of isosurfaces, so this view is defined as a set of ESIP surfaces

• We need to describe this resource isosurfaces in terms of the ontology:

Isosurfaces isa SurfacesIsosurfaces isa GeometryIsosurfaces isa AtomicViewIsosurfaces isa View

Isosurfaces description

Note: VisKo relies on Resource Document Framework (RDF) for encoding statements

Encoding Velocity Model Operator

• A contouring operator generated the isosurfaces• The contouring operator– operated on data in format 3DImageData– generated the view isosurfaces– output plot in format PolyData

contouringOperator isa MappercontouringOperator operatesOn 3DImageDatacontouringOperator transformsTo PolyDatacontouringOperator mapsTo isosurfaces

vtkContourFilter description

Note: contouring operator is conceptual and cannot be executed

Encoding Velocity Model Service

• The contouring operator was implemented by the VTKContourFilter service

vtkContourFilter isa ServicevtkContourFilter implements contouringvtkContourFilter supportedBy VTK

vtkContourFilter hasInput contourSpacingvtkContourFilter hasInput numberOfContoursvtkContourFilter hasGrounding wsdlGrounding

vtkContourFilterdescription

VisKo-Service

OWL-S

Executable VisKo service implements operator contouring

Presentation Overview

1. Toolkits and Visualization Pipelines2. Visualization Query3. Automated Generation of Visualization pipelines4. Ontological Description of Visualization Pipelines5. Execution of Visualization Pipelines6. Giovanni and VisKo

Pipeline Execution

• VisKo infrastructure is also responsible for executing the derived pipelines

• Pipeline Execution Requirements– All operators must have a Web service implementation– Any operator that does not have a service implementation

cannot be executed, and neither can its parent pipeline

Visualization Language Layer Cake• VisKo language does not express execution details• VisKo is therefore is built on other

languages/standards that specialize in execution of services and pipelines

WSDL

GMT VTK IDL

OWL-S

VisKo

Executable Toolkit Operators

Web Services

Semantic Web Services

Visualization Language

SEMANTIC

SYNTACTIC

PROGRAMINGLANGUAGE

Execution layers

Presentation Overview

1. Toolkits and Visualization Pipelines2. Visualization Query3. Automated Generation of Visualization pipelines4. Ontological Description of Visualization Pipelines5. Execution of Visualization Pipelines6. Giovanni and VisKo

Giovanni vs VisKo

Giovanni VisKo Capability

Visualization

Configurable views

Pipeline (i.e., workflow)

Multiple Toolkits

Provenance Capture

Dimension Compression

Dynamic Workflow

Plug and play

Automatically Adaptable

Giovanni From GrADS to NCL

• Had to rework Giovanni when NCL was introduced• Workflow may have needed to be modified:– Added operators to convert to new format required by

new operator– Configure or change actor to work with protocol of new

visualization operator– Input new parameters to the visualization module

Giovanni-VisKo

• VisKo could mediate some issues associated with:– Adding new visualization capabilities– Modifying existing capabilities– Workflow versioning

• Underlying toolkits and pipelines will constantly change but the visualization query should remain the same

• So the interface to VisKo will never change

Giovanni-VisKo (1)

• Giovanni’s Kepler workflow is performing two functions:– Scientific workflow: bias correction, filtering, gridding– Visualization: possibly converting to format that NCL

renderer can process

• If Giovanni workflow could be partitioned into two sub workflows:– It could focus only on the scientific data processing– And delegate the visualization responsibilities to VisKo

Thanks for listening!

VisKo Ontologies

VisKo-View Ontology

• Defines different views and view properties

infovis Propertiesscivis

Imported from ESIP Data-type ontology

ESIP Views

• Defines geometries of scientific data with different dimensionality of spatial components

VisKo-Operator Ontology

• Defines different kinds of visualization toolkit operators

Operators

Format

Imported from PML-P provenance ontologyImported from VisKo-View

Note: operator concepts are conceptual and cannot be executed

VisKo-Service Ontology

• Defines relations between an operator and OWL-S• Imports OWL-S Service (service)• Imports OWL-S Process (parameters)

Cardinality: an operator can be implemented by many services

Cardinality > 1

Data type

Pipeline Procedure and Rules

Pipeline Composition Procedure

• Non-logic based description of algorithm

Using our knowledge about visualization operators:1. Find operator A that works on format of input data2. Find operator B that works on format output from

operator A3. Chain A and B4. A B; Repeat from 2 until output format of B can be

ingested into operator that generates requested view

Composition Rules

• Rules that derive pipelines are applied to– statements about the visualization toolkit operators– Statements asserted in the query

• The composition rules are encoded in OWL2– Inverse properties– Property chains– Property transitivity

OWL Composition Rules

canBeTransformedTo(?Fmt1, ?Fmt2) :-isOperatedOnBy(Fmt1, A),transformsTo(A, Fmt2).

A

Legend:

Asserted PropertiesoperatesOntransformsTo

Inferred PropertiesisOperatedOnBycanBeTransformedTo

B

Fmt1

Fmt2

Fmt3

OWL Transitive Property Chain

Transformer

isOperatedOnBy(?Fmt1, ?A) :- operatesOn(A, Fmt1).

OWL Inverse Property

OWL-S and WSDL

The Role of OWL-S

• Visualization toolkits:– Have different protocols for interfacing with operators– Are implemented as executable, C/C++ libs, Java libs etc.

• OWL-S adds a layer over the different protocols and “levels the playing field” so to speak…

VTK(C++) GMT(.exe)

OWL-S

VisKo interfaces with only OWL-S

WSDL

Declarative

Imperative

The Role of WSDL

• Web Service Description Language is an XML format for describing service endpoints….syntactically

• OWL-S to bridges semantic/syntactic gap• Toolkit operators must be exposed as WSDL services

to be considered in the VisKo framework

WSDL Document