Visualizing and Tracking Features in 3D Time-Varying...
Transcript of Visualizing and Tracking Features in 3D Time-Varying...
![Page 1: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/1.jpg)
Feature Extraction & Tracking: What next?
Deborah SilverProf, Department of Electrical and Computer EngineeringRutgers, The State University of New Jersey
http://www.caip.rutgers.edu/vizlab.html
August 2009
![Page 2: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/2.jpg)
Space/Time reduction
As the datasets become larger and larger, it becomes physically impossible to do in-depth discovery of all of the data. Filtering techniques are necessary to help the scientist focus on regions of interest in the space/time domain.
![Page 3: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/3.jpg)
Data Reduction
Filter/pre-process/reduce
visualization
![Page 4: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/4.jpg)
Effective Filtering
To Create Effective Filtering + Visualizations of Massive
3D+ Time Varying Simulation Data
Feature Extraction & Tracking
![Page 5: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/5.jpg)
Motivation
terabytes – very difficult to look through all of them – need a better way to search
![Page 6: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/6.jpg)
Example-3D Event Querying
Automatically find interesting eventsFollow Topological changesClassify eventsSearch eventsVortex Reconnection
??
![Page 7: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/7.jpg)
What is tracking?
Following “features” over time“Features” can be anything --- defined as coherent blobs/objects meeting certain conditions
![Page 8: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/8.jpg)
Weather Simulation
Resolution 42x57x30, 37 time steps
Data courtesy of Dr. YanChing Zhang at EPA
Isosurface of Cloud Water at threshold =0.00029
Pseudo-spectral Simulation.Isosurface of vorticity magnitude at 48% of maximum1283, 100 timestepsn (shown)
![Page 9: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/9.jpg)
Difficulties:
Size – too much data Clutter (Visual)Quantification: measurements Querying capabilities:
How many regions are there?Where are the big regions?Is “XXXX” present?How does this compare to a different simulation?
Classification: store in a database
![Page 10: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/10.jpg)
Feature-based Process Model
![Page 11: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/11.jpg)
Major Components
Feature Extraction Define the features of interest. Domain dependent. Pre-defined or interactive.
Feature TrackingAutomatically correlate extracted regions from one dataset to the next
Quantification / Measurements for extraction & tracking.
--> BETTER VISUALIZATION
![Page 12: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/12.jpg)
Features
Basic definitionRegions of interest consisting of connected nodes satisfying some criteria (e.g. threshold interval, topological specification)
Each domain has its own definitionVolume intervals [G95]Segmentation [S95]Selective Visualization [W95]Domain specific: Shock waves, Vortex Cores, Eddies, Medical ...
![Page 13: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/13.jpg)
Feature Abstraction
Abstract feature using a “reduced” representation objectCompression of geometryEncapsulation of idea --- non-photorealistic renderingReduced modeling
![Page 14: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/14.jpg)
Abstraction: Data Reduction
Local maxima:
Abstraction using skeletons
Abstraction using ellipsoids
![Page 15: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/15.jpg)
Feature Tracking
Automatically correlate extracted regions from one dataset to the next
Assumption:Sufficient Sampling Frequency such that corresponding features overlap in space.
![Page 16: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/16.jpg)
Tracking
ContinuationBifurcationAmalgamationDissipationCreation
![Page 17: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/17.jpg)
Observations
•Continuation: if feature Oi
A corresponds to Oi+1
B , then OiA
overlaps with Oi+1B .
•Bifurcation or Amalgamation: if a feature splits into a group of N objects, then all Oi
A in N overlap with Oi+1
B
Continuation Bifurcation/Amalgamation
Time 1 Time 2
![Page 18: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/18.jpg)
Visualization Paradigm
DAGEnhanced surface renderingEnhanced volume renderingFeature isolationTrace and trajectoryFeature Juxtaposition
![Page 19: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/19.jpg)
Enhanced Surface Rendering
![Page 20: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/20.jpg)
Dr. Zhang, EPAWeather Simulation
Graphs for Real-time monitoring
![Page 21: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/21.jpg)
Feature Isolation/Quantification
Standard Isosurface
Forward Isolation
Backward Isolation
Quantification
![Page 22: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/22.jpg)
Juxtaposition
![Page 23: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/23.jpg)
Trace and Trajectory
![Page 24: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/24.jpg)
Event Graph Visualization
![Page 25: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/25.jpg)
Tracking issues for Ultra-scale Visualization
Post-processing tracking too slow, data too large tracking while simulation is progressing
Feature extraction must be presetMechanism to change thresholds etc..Quantities extracted
Real time steeringCan be used for simulation (feedback to simulation)Multiresolution dataDatabase classification for discovery
Challenges- robust code, distributed code, standard in vis packages
![Page 26: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/26.jpg)
Simulation DATA
Real-time monitoring of pre-processed data
Post-Processing in-depth discovery
Tracking can be done as part of a pre- or post-processing of the data.
![Page 27: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/27.jpg)
Distributed feature extraction and tracking. Each processor computes a partial extraction and tracking using ghost communications. The full solution is then merged.
Feature tracking from ti to ti+1
Partial merge across processor boundaries
![Page 28: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/28.jpg)
Timeline for parallel feature extraction and tracking system
Distributed Feature Tracking
![Page 29: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/29.jpg)
AMR datasets: Adaptive Mesh Refinement
2D AMR
Regridding from ti to ti+1
Challenges…
Level 1 Level 2
A splits into two pieces in Level 2: C2 and C3
![Page 30: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/30.jpg)
Visualization Updates
Feature TrackingThe feature extraction and tracking system, previously implemented for AVS/Express, has been ported to the VisIt visualization tool. In the VisIt version of the feature tracking system the computations have been decoupled from the visualization of the results, to allow faster and more flexible viewing.
![Page 31: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/31.jpg)
Feature Tracking within Visithttp://www.eden.rutgers.edu/~anaveen/VisIt/VisIt.html
![Page 32: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/32.jpg)
Feature Tracking pipeline
![Page 33: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/33.jpg)
![Page 34: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/34.jpg)
What next?
![Page 35: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/35.jpg)
Event Classification
For massive simulations Database querying functions (pre-processed)Event tracking
![Page 36: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/36.jpg)
Example-3D Event searching
Automatically find interesting eventsFollow Topological changesClassify eventsSearch eventsVortex Reconnection
??
![Page 37: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/37.jpg)
Data is too large to look at
Similar to large databases – need to query the data“higher level queries”Isolate when a particular event occursShow when a behavior happens or is about to happenNeed semantics to specify the query
![Page 38: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/38.jpg)
Feature & Event classification for Fusion-Feature Based Techniques to characterize and catalogue interesting phenomena
(Plasma specific)
BlobsFilamentsAvaloidsStriationsBurstsRadial streamers IPO (Intermittent plasma Objects)Holes (opposite of blobs, density rarefactions)Chaotic Field line regions
(CFD-general)bubble holeblast wave packetblobcloud patchcritical pt. pointeddy ring
(Plasma specific)
Spin/wake (blobs)Zonal flowsFlow shearsRotation
(CFD general)
advect swirlentangle transportdisperse windflowhopmigratestream
(Plasma specific)
Coalesce (blobs)Breakup (blobs)
(CFD General)
accreteaggregatealignbind bifurcateburstcollapse
Objects/Features Move Interact ----Come Together/Apart
Favor rollfilament separatrixfinger spikegyres spiralhairpin striationhelix vortex
condense disassembledisruptfingerfissionfocusfoldfusepair
roll-upplowreflectscatterspikesplitstriatestripwind about
WigglesFlux tubesLoss Cone
![Page 39: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/39.jpg)
Objects Move Interact ----Come Together/Apart
Combustion specific Combustion specific Combustion specific
KernalFlame (premixed, part, diffusion,edge)ShockwaveFuel JetStochiometric LineRegion of chemical reactionRegion of Flow
burn curve convect diffuse ignite
quenchpercolate propagatereignitereactstrain
flame-wall interactionsmerge annihilate upstream annihilate downstream
CFD General
accreteaggregatealignbind bifurcatebreakupburstcollapse
condense disassembledisruptfingerfissionfocusfoldfusepair
roll-upplowreflectscatterspikesplitstriatestripwind about
CFD-general CFD-general
bubbleblast waveblobcloud critical pt.eddy
holepacketpatchpintring
favorfilamentfingergyrehairpinhelix
rollseparatrixspikespiralstriationvortex
advectentangledisperseflowhopmigrate
streamswirlttransportwind
Characterization of events for combustion
![Page 40: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/40.jpg)
Challenges
MOST IMPORTANT: Usable extraction/tracking code – either libraries or end applications to allow scientists to try it and not just visualization experts
![Page 41: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/41.jpg)
Challenges
Track neighborhoods, not just atomic featuresCharacterize events – sequences of actions not just the “simple” tracking actionsData structure to store events (multimedia data structures) Each domain has its own events – are there commonalities so that a generic system could be developed?
![Page 42: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/42.jpg)
Challenges
New meta-data is created which also creates new visualization issuesFeature data baseEvent comparisons, event monitoring, controlWhat is the best way to present tracking information (perceptual issues).
![Page 43: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/43.jpg)
Why Feature TrackingReduce the Size of DataReduce ComplexityProvide QuantificationEnhance VisualizationFeature based QUERYINGFacilitate Event SearchingHelp with code comparisons, help with simulation/observation analysis. Can only be done on a higher level comparion
![Page 44: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/44.jpg)
CPES Visualization
![Page 45: Visualizing and Tracking Features in 3D Time-Varying Datasetscscads.rice.edu/deborah-silver-cscads-2009.pdf · Post-processing tracking too slow, data too large Æ tracking while](https://reader034.fdocuments.net/reader034/viewer/2022042300/5ecaf75b31e6bc613a32fe07/html5/thumbnails/45.jpg)
Thank You!!
http://www.caip.rutgers.edu/vizlab.html