Interactive Interface Design for Scalable Large Multivariate Volume Visualization

Transcription

1 Interactive Interface Design for Scalable Large Multivariate Volume Visualization Xiaoru Yuan Key Laboratory on Machine Perception, MOE School of EECS, Peking University Nov. 13 th 2011

2 Outline Motivation Multivariate Volume Transfer Function Design Parallel coordinates & MDS Scattering Points in Parallel coordinates Parallel extension of the TF design Scalable Pivot MDS Adaptive Continuous Parallel Coordinates 2

3 High Dimensional/Multivariate Data Set Isabel Hurricane QVAPOR QCLOUD Pressure Speed

4 Transfer Functions Transfer functions map the voxels values to colors and opacities, generating insightful results. 1D TF (Intensity) 2D TF (Intensity vs. Gradient Magnitude)

5 Temperature Transfer Functions Multivariate TFs for multi-modal data QVAPOR QCLOUD Pressure Speed Pressure

6 Methods of Visualizing Multi-dimensional Data Scatterplot Matrix Star Glyphs Chernoff Faces Multidimensional Scaling (MDS) Parallel Coordinates etc. 6

7 Multidimensional Scaling 7

8 Parallel Coordinates To represent N dimensional data Set N vertical axes in parallel Put data to intersects on corresponding axes Connect intersects 8

9 Data Exploration with PC 9 [Yuan et al. TVCG 2009]

10 Multivariate Visualization with Parallel Coordinates 10 [Jones et al. 2008]

11 Line vs. Point Representation 11

12 Line vs. Point Representation 12

13 Line vs. Point Representation 13

14 Data Exploration with SPPC 14 [Yuan et al. TVCG 2009]

15 Data Exploration with SPPC 15 [Yuan et al. TVCG 2009]

16 High Dimensional Transfer Function Design Key issue in multivariate TF design: Identifying features in multi-dimensional space Multidimensional data visualization in InfoVis community applied on TF design: Parallel coordinates plot (PCP), which keeps information on each dimension Multidimensional scaling (MDS), which works better on multidimensional feature identification PCP MDS

17 High Dimensional Transfer Function Design Integrate MDS into PCP to facilitate multivariate TF design. Avoids context jumps between polyline and point regions when exploring data clusters Provides multiple perspective views upon the data, supporting linked queries [Guo, Xiao &Yuan, PacificVis 2011]

18 Design Multivariate TFs with the Proposed System

19 Volume Rendering with Sketch Feedbacks

20 System Pipeline (Non-parallel version) Pre-Processing Reconstruction

21 Transfer Function User Interface

22 Transfer Function User Interface User interactions Brushing on axes Lasso on points Magic wand on points Lasso Tool Magic Wand Tool

23 PCP and MDS Generation Pre-Processing Subroutines: Embedding MDS Plot into Continuous PCP Generation of Weight-Adjustable MDS Plot Adaptive Rendering of Continuous PCP

24 MDS Plot Pivot MDS [Brandes and Pich 2007] Low storage and low computational complexity MDS works better for feature identification and selection

25 MDS Plot Hierarchical adaptive sampling can be exploited to reduce the data amount to progressively reaching optimal

26 MDS Plot Metric adjustable MDS Allows user to define different impacts from the dimensions on MDS layouts

27 TF Construction Gaussian Mixture Model (GMM) Use several Gaussian blobs to fit the distribution of user selected clusters Reconstruction

28 Hurricane Isabel

29 Hurricane Isabel Red part (hurricane eye): low pressure, medium temperature, lower QCLOUD, high wind speed

30 Hurricane Isabel Yellow part: higher pressure lower wind speed

31 Hurricane Isabel 31

32 System Pipeline Pre-Processing Reconstruction

33 System Pipeline Parallel Environment

34 Adaptive Continuous PCP 5% 10% 15%

35 Hidden Features Adaptive Continuous PCP The continuous PCP is a HDRI (High Dynamic Range Image). Features may be hidden without proper mapping strategy. A logarithm tone-mapping is utilized to enhance the small features Pseudo Color Linear Logarithm

36 Pivot MDS 36

37 Pivot MDS Parallization Computation of squared distance matrix Δ Double centered sub-matrix C Inner product CTC Eigensolver 37

38 Scalable Pivot MDS

39 Performance Parallel Multivariate Volume Rendering

40 Performance Parallel MDS Projection

41 Performance Parallel PCP Rendering

42 Integrated System Interface 42

43 VisIt Plug-in (ongoing) 43

44 Related Publications H. Guo, H. Xiao, and X. Yuan. Multi-dimensional transfer function design based on flexible dimension projection embedded in parallel coordinates. In Proceedings of IEEE Pacific Visualization Symposium 2011, pages 19 26, H. Guo, H. Xiao, and X. Yuan. Scalable Multivariate Volume Visualization and Analysis based on Dimension Projection and Parallel Coordinates. IEEE Transactions on Visualization and Computer Graphics. Under revision. 44

45 Acknowledgement Students Hanqi Guo, Xiao He Collaborator CNIC CAS (parallel computing environment) Funds NSFC , NSFC Beijing NSFC Project 2010AA Chinese Ministry of Education Key Project No

46