electronic visualization laboratory university of illinois at chicago “time-critical...

11
Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware” by Xinyue Li, Han-Wei Shen Presented by: Allan Spale, CAVERN Viz Workshop, May 2004

Upload: maud-singleton

Post on 18-Jan-2018

217 views

Category:

Documents


0 download

DESCRIPTION

Electronic Visualization Laboratory University of Illinois at Chicago Traditional LOD Selection Methods Static heuristics –Use constant criteria View angle, screen coverage, object distance –Difficult to use because of varying heuristics and imprecision of rendering time Inter-frame feedback –Calculate change in rendering time between frames –Problem of oscillation in rendering time between frames Global optimization –Using a user-specified time frame, maximize quality of rendering and minimize rendering cost –Problem to use since pipeline performance numbers are difficult to obtain with many dynamic factors in action

TRANSCRIPT

Page 1: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

“Time-Critical Multiresolution Volume Rendering using

3D Texture Mapping Hardware”by

Xinyue Li, Han-Wei Shen

Presented by:Allan Spale, CAVERN Viz Workshop, May 2004

Page 2: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Overview

• LOD selection algorithm for multiresolution hierarchical volume rendering– LOD automatically selected and conforms to

user’s rendering criteria • Frame rate kept within 10% of user’s requested frame

rate– Rendering metrics collected during runtime and

used in a intra-frame predictive scheme for LOD selection

• Uses 3D texture mapping hardware

Page 3: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Traditional LOD Selection Methods• Static heuristics

– Use constant criteria• View angle, screen coverage, object distance

– Difficult to use because of varying heuristics and imprecision of rendering time

• Inter-frame feedback– Calculate change in rendering time between frames– Problem of oscillation in rendering time between frames

• Global optimization– Using a user-specified time frame, maximize quality of rendering

and minimize rendering cost– Problem to use since pipeline performance numbers are difficult

to obtain with many dynamic factors in action

Page 4: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Multiresolution Volume Redering• Hierarchical volume generation using algorithm by Weiler et

al. [paper reference 2] – Generate lower resolution by averaging every 2 x 2 x 2 voxels– Reduce seams between between nearby subvolumes by copying

low-res volume data points on boundaries to high-res volume– Subvolumes created according to complexity in each space…more

complexity more subdivision• Rendering using 3D texture mapping hardware

– Slice polygons perpendicular to view vector– Use OpenGL to shade and blend the polygons– At program initialization, create OpenGL texture object for each

LOD subvolume– Rendering involves binding a texture object based on LOD data

• Encourages reuse of existing textures

Page 5: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Algorithm: Intra-Frame Prediction Reaction

• Subvolumes are rendered in sequential order, front-to-back– Each choice about subvolume

made at time of rendering• Obtain performance stats of

previously rendered subvolumes• Track time spent and importance

values (determined by the user) for to-be-rendered subvolumes

• Based on relative importance, allocate rendering time slice and choose LOD matching this constraint

– Repeat steps until all subvolumes are rendered

Page 6: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Algorithm: Rendering Time Prediction

• Equations T = tloading + tprocessing * nslices

tloading : LOD volume texture loading time (do not include the value in tprocessing )

tprocessing : With certain assumptions, obtain the time spent rendering slices for an LOD divided by the number of slices rendered so far, nslices

tprocessing=(tavg_voxel * nvoxel) * (1-R) tavg_voxel : Average loading time of each

LOD voxelnvoxel : Voxels in the LODR : Boolean value

1: Texture in memory0: Texture not in memory

Page 7: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Algorithm: Time Budget Allocation

• Importance metrics affect a subvolume’s rendering time ti = Tb * (Ii / j=1..k Ij)

Tb: time remaining; k: number of remaining subvolumes

I: importance value – i= current, j= remaining values; Below are importance criteria (properties taking more time are listed)

Maximum opacity: more opaqueDistance to viewpoint: closerProjection area: higher valueGaze distance: (distance from center

of gaze area to center of projected area’s subvolume) closer to gaze area

Page 8: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Algorithm: LOD Selection, Temoral Coherence Consideration

• LOD Selection– For every LOD texture, check if texture is

in memory– Based on rendering history, calculate the

average rendering time per slice– Based on the performance model,

determine the predicted rendering time for each LOD subvolume

– Select the LOD that minimizes the difference between predicted rendering time and budgeted rendering time

• Temporal Coherence Consideration– Avoid flicker with frequent LOD changes

• Each change in viewing direction or gaze area above a threshold results in a recalc of each subarea’s importance value

Page 9: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Results:Frame Rate, LOD Selection

• (Figure 2) Rendering style; time critical algorithm within 10% of target frame rate– Identical viewing parameters; 200 frames (first and last 50 frames:

different scaling factors, middle 100 frames: different viewing angles

• (Table 2) Processor time for LOD selection is quite small

Page 10: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Results: Time Budgeted, Time-Based Gaze-Directed Rendering

• (Table 4) When considering the importance factors, rendering time is noticeably adjusted for low and high categories with respect to ignoring importance factors

• (Table 6) Subvolumes with a small gaze distance (nearest) receive more rendering time with respect to ignoring gaze distance

Page 11: Electronic Visualization Laboratory University of Illinois at Chicago “Time-Critical Multiresolution Volume Rendering using 3D Texture Mapping Hardware”

Electronic Visualization Laboratory University of Illinois at Chicago

Summary• Highlights

– Multiresolution visualization using importance factors• Importance factors assist in an automatic LOD selection

– Supports texture mapping hardware• Pros

– Can maintain a steady frame-rate– Subvolumes divided according to complexity and individually

rendered at different LODs– Control algorithm has very minimal overhead

• Cons– Current work done on “small” datasets (done using single

PC?)– Little difference between “low” and “medium” importance