Interactive Information Visualization of a Million ItemsJean-Daniel Fekete Catherine Plaisant

Human Computer Interaction Laboratory

University of Maryland

http://www.cs.umd.edu/hcil

Originally Presented at InfoVis 2002

Reviewed for CS 526 by John T. Bell

Project Goals

• To display very large numbers of data items

• Each item to be displayed as a separate atomic object, not aggregated.

• Visualization to be cognitively perceptual.

• Provide continuous interaction and smooth animation when changing viewing parameters or view types.

Classic Visualization Techniques

Overlapping Space-Filling( Scatter Plot ) ( Treemap )

TreeMap of a Million Items

TreeMap of a UNIX Filesystem

Human Perception:Pre-Attentive Graphical Features

Nelson

Johnson

Yu

Buy

Bush

Bell

Theys, M.

Theys, C.

Pre-Attentive Features ( Healy [ 11 ] )

• orientation

• length

• width

• size

• curvature

• number

• terminators

• intersection

• closure

• color (hue)

• intensity

• flicker

• direction of motion

• binocular luster

• stereo depth

• 3D depth cues

• lighting direction

Interactive Techniques ( Ahlberg [ 1 ] )

• Overview, zoom in, filter, details

• Display query & results

• Rapid, incremental, & reversible control

• Selection by pointing, not typing

• Immediate & continuous feedback

• Techniques for large data sets include[ space time ] multiplexing & space deformations ( aggregation, sampling, LOD )

Technical Constraints

• 1600 x 1200 displays -> 1.92 Million pixels

• Refresh rate of 10 fps requires special techniques. ( Graphics cards at 15M tps, achievable in practice only with Δ strips. )

• This work done with NVidia GeForce3 cards, 2GHz computers, OpenGL

• Interactive speeds achieved by cleverly offloading rendering tasks to the GPU

Individual Item Display

• Items displayed as shaded quads, w/o borders

• Information attributes include size, position, color ( categorical or value based ), and intensity ( value based )

• Color indicated using one-dimensional texture indexes. ( One byte per vertex, not 3 or 4 )

• Data per vertex: X, Y, Z, and S ( text. Index )

Illustration of Shaded Rectangles

Synthetic Overlap

• Stencil Buffer can be accessed to determine overlap counts and/or filtering accordingly:

Transparency & Stereo Vision

• Transparency can be useful in overlap situations, BUT it interferes with pre-attentive processing by blending colors.

• Therefore only useful when dynamically adjustable for exploring overlap information.

• Stereo vision also of limited usefulness with overlap exploration, for the same reasons.

Animation and Interaction• Data exploration often involves changing views - This

can cause problems correlating information in different views.

• Flipping can be used when geometry is fixed.

• Linear interpolation works when layout fixed.

• Otherwise interpolate in two stages: Postion first, then size. ( Treemaps can be squarified or slice-and-dice. )

• Texture maps can speed up the process, by warping a square area as a whole unit.

Animation of Property Changes

Animating View Changes

Implementing Dynamic Queries• Used to dynamically filter the data based on a range of

values for a particular attribute.• Items are stored as display lists on the GPU• Z-coordinate is mapped to changing attribute• Near and far clipping planes are adjusted based on slider

values, and the GPU solves the problem of which items to display.

• For PCs with limited video RAM, use OpenGL points of varying sizes instead of quads ( NVidia Vertex programs extension. )

Performance

• 23,000 lines of C++, reads XML or directory

• NVidia GeForce 3 on 2GHz Pentium and 3Dlab Wildcat 5110 on dual 1.7 GHz Pentium

• Sustainable performance ~2.5 M Quads / sec

• 10 fps treemap animations for any treemap, through the use of texture maps.

• 3 / 6 fps scatterplots, for animations / queries.

Conclusions & Future Work

• 1M items have been displayed on 1600x1200, interactively without aggregation.

• New techniques have been developed to take advantage of modern GPU capabilities.

• Future data explorations include U MD catalog and circulation data and U.S. Census data, involving domain experts in user testing.

Web Site, Downloads, Movies

• Http://www.cs.umd.edu/hcil/millionvis

References - General• Jean-Daniel Fekete & Catherine Plaisant, “Interacive

Information Visualization of a Million Items”, InfoVis 2002.

• [1] Ahlberg et al., “Visual Information Seeking: Tight Coupling of Dynamic Query Filters with Starfield Display”, Human Factors in Computing Systems, 1994.

• [11] Healy et al., “Visualizing Real-Time Multivariate Data using Preattentive Processing”, ACM Transactions on Modeling and Computer Simulation, 1995.

References - Treemaps

• [4] Bederson et al., “Ordered and Quantum Treemaps: Making Effective Use of 2D Space to Display Hierarchies”, ACM Transactions on Computer Graphics.

• [13] Johnson et al., “Tree-maps: A Space-Filling Approach to the Visualization of Hierarchical Information Structures”, IEEE Visualization ‘91.