Debates: Overview+Detail vs. Focus+Context 2-D vs. 3-D cs5984: Information Visualization Chris North
Multi-Dimensional data• Variety of studies:
• Merwin, Wickens, Boyle, Boyer, Hollands, Barfield
• data plots, air traffic control, …
• E.g: 2 2D views vs. 1 3D view vs. 2D+color
Results
• Hard to make sense of it all…
• In general:• 2D: better accuracy, efficiency
• 3D: better overview, trends
• Architects know this!• Orthographic 3d for overview
• Plan, profile, section (?) for exact measurements
Ware et al.• Node+link graph structures in 3D space• 4 versions:
• Static 2D = 3D graph projected onto 2D, static
• Stereo 3D
• Rotation 3D (head-coupled)
• Stereo+Rotation 3D
• Task: path finding• Exist path of length 2 between 2 highlighted nodes?
Results• Error rates different, Response times same
• Static 2D worst
• Stereo 3D: 1.6x better
• Rotation 3D: 2.2x better
• Stereo+Rot: 3.0x better
• Method of rotation not important
So…?
• Static 3D beat static 2D• 3D good for revealing complex structure?
• Better to see things that are 3D in 3D! (Hubona)
• 2D = 3D - stereo - rotation• Of course it’s the worst!
• What about 2D interaction, alternative layouts?• Springs, aggregation, focal nodes, hierarchization,…
Modjeska
• Hierarchical data in VR
• 2 UIs:• 3D fly through (6 dof, VRML) similar to SGI FSN
• 2D map-view (top view of VR) with zoom/pan like Pad++
Results
• Number of search targets found:• 3D: 11.0
• 2D: 13.6 (sig)
• 3D navigation cumbersome
• 3D version is basically 2D!
Carr et al.• 3D visualizations for
Hierarchical data:• Landscape: like 3D in Modjeska
• InfoCube: 3D containment, boxes in boxes, transparency
• CamTree: like ConeTree
• VRML 6 dof navigation only
• Tasks:• Search
• Count
• Compare
Results
• 3 Tasks: Search, Count, Compare
• Best to worst (perf time & errors):• Landscape
• CamTree
• InfoCube
• Landscape: users used as 2D!• Bird’s eye, then zoom in
So…?
• 3D (6 dof) navigation cumbersome
• Custom interaction techniques more important• When doing 3D Vis, don’t just do 6 dof VR
• E.g. ConeTree directory rotations, pruning, DQ, etc.
• VR can beat PR! (Physical Reality)
• Need overview map for 3D • Disorientation problem
• e.g. Harmony web browser
Conflicting Results?
• Design of 3D navigation controls• Head tracking (natural) vs. control panel (unnatural)
• “Look at” vs. “Be in”• Fish-tank VR vs. Immersive VR
• Fish-tank: 2 dof, never lost, easier
• Immersive: 6 dof, disorientation, harder
• Multi-D vs. Graph vs. Tree• Tree can be done well in 2D
Summary
• Interaction design more important than 2D/3D• Can make good displays in 2D or 3D
• Interaction makes or breaks
• Currently: 2D interaction designs more advanced» Fisheye, F+C, O+D, zoom/pan,…
• Fish-tank 3D: • Promising results for visualizing complex structure
• E.g: Networks, biology molecules
To Do
• Compare good 2D Vis. to good 3D Vis.• E.g: HyperbolicTree vs. ConeTree
• Theory• Cost structure of information access (Card et al.)
• Calculate clicks/time to access info
• What about Immersive VR?
• Tool builders + evaluators• Builders don’t test their good tools
• Evaluators do good tests on lame tools
Project Presentations• Dynamic Data Visualization
» Umer, Dilshad, Satyajit
• Multi-Dimensional Parameter Space Visualization » Ravi, Prasuna, Ashwini, Vijay
• Web Snap» Sanjini, Joy, Aarthi
• Data Structure Visualization Evaluation» Priya, Gowri, Fanye, Aejaaz
• Data Structure Visualization Tool» Sumithra, Luhui, Shumei (and Matt, Sam)
• Biotech Visualization» Margaret, Josh, Matt, Yuying
• Chat Log Visualization » Marcus, Marty, Purvi
• Menu Visualization» Chris
• Data Density and Distraction Evaluation» Maulik, Ajay, Denzil
Thursday,April 26
Tuesday,May 1