Interactive View-Driven Evenly Spaced Streamline PlacementZhanping Liu Robert J. Moorhead II

Visualization Analysis and Imaging LabHigh Performance Computing CollaboratoryMississippi State UniversityIS & T / SPIE EI-VDA 2008

Outline

Introduction

Introduction Evenly Spaced Streamlines (ESS)

Introduction Evenly Spaced Streamlines (ESS)

Introduction Evenly Spaced Streamlines (ESS)

Introduction Evenly Spaced Streamlines (ESS)

Introduction

Introduction

Introduction

IVDESSwhether a streamline is further advectedor immediately terminated in physical spaceis governed by the status (accepted/rejected)of the newly generated pointthe projection of each streamline pointand the associated view-space samplesundergo inter-sample distance controlto achieve inter-line distance control

IVDESS The Pipeline

IVDESS The Pipeline

IVDESS Temporally Coherent Seeding Strategy IVDESS employs an inter-frame physical-space seeding scheme on top of TISS to constitute a The inter-frame physical-space seeding scheme maintains temporal coherence by reusing and lengthening the streamlines of the previous frame under normal density control in the current frame Temporally Coherent Seeding Strategy (TCSS) physical-space seeding prior to view-space seeding

IVDESS TCSS

IVDESS TCSS Efficient Greedy Non-split Streamline Reuse+Lengthening Point projection and segment sampling continue until any sample (I1) fails to pass inter-sample distance check an intermediate segment sample from line-view clipping (seed S out of view)

Check view-space length to decide if the streamline needs saving

IVDESS View-Sensitive Streamline Representation

IVDESS View-Sensitive Streamline RepresentationJaggy lines emerge when unintended unprojection points (due to numerical error) of the current frame are reused in the subsequent frames to lengthen the streamlines

Results Test dataset a 468337 2D flow field of the Northeast Pacific ocean

Results Play the IVDESS-TISS movie! Play the IVDESS-TCSS movie!

Results

Resultsstreamline reuse percentage for each TCSS frameThe high percentages demonstrate the effectiveness of the greedy non-split streamline reuse+lengthening scheme adopted in TCSS.

Results

ResultsThe interactive and nearly constant frame rates of TCSS indicate that IVDESS-TCSS (IVDESS) is well suited for coherent flow exploration.

Conclusions IVDESS is a physically non-uniform but visually uniform representation of planar or curved surface flows in a perspective-view setting IVDESS divides the view-dependent uniform placement process into physical-space flow integration & view-space streamline density control A projection-unprojection pair is used via off-screening surface rendering to link the two spaces Greedy but efficient non-split streamline reuse+lengthening is an inter-frame physical-space seeding scheme that is adopted on top of an intra-frame view-space seeding method to constitute a hybrid-space multi-level seeding mechanism Temporally Coherent Seeding Strategy A view-sensitive streamline representation is used to support thorough reuse+lengthening while guaranteeing proper rendering IVDESS is well suited for coherent level-of-detail 3D exploration of large complex flows at interactive frame rates without either pre-processing or GPU support on a nowadays low-end PC

Conclusions DoD HPCVI Program Dr. David Kao Anonymous reviewers Acknowledgments to enhance the current version of IVDESS in support of flows on curvilinear grids and unstructured grids to investigate adaptive depth selection issues in an effort to extend IVDESS for explorative visualization of volume flows Future WorkIS & T / SPIE EI-VDA 2008 Thank you!Any questions?

I am Zhanping Liu, from Mississippi State University.

I would like to present our work on interactive view-driven evenly-spaced streamline placement.First I will give a brief introduction to evenly-spaced streamlines and previous work related to ours.

Then I will talk in detail about our IVDESS algorithm.

Finally I will show you some results and summarize our work with future plans.There have been a lot of flow visualization methods including geometry-based and texture-based.

Texture-based techniques are USUALLY effective in visualizing 2D flows.However, sometimes they may be ineffective for surface and volume flows due to some perception problems.

On the other hand, streamlines remain one of the most straightforward and fastest techniques for 3D scenarios.But without a sophisticated streamline placement strategy,

the result tends to be an incomplete coarse view

or a global but cluttered image.A GOOD placement of evenly-spaced streamlines may provide a visually pleasing and informative pattern to facilitate mental reconstruction of the flow.

Here we use ESS for evenly spaced streamlines.To extend ESS for 3D scenarios,

we need to address the foreshortening effect because streamlines evenly spaced in 3D physical space may not visually retain the uniformity when projected to 2D view space through perspective projection.

Here are two examples showing the problems caused the foreshortening effect.

In addition, we need to consider inter-frame transition and practical applicability.

Existing ESS algorithms can be roughly categorized into image-guided and sample-based.

The former take any streamline placement as a binary-valued image and employ low-pass image filtering in an iterative placement refinement process.

Sample-based algorithms use inter-sample distance control to approximate inter-line distance control.

Specifically, distance checking is performed on each newly generated sample of a streamline against other existing samples

to determine whether or not the distance is less than a threshold value.There have been several ESS algorithms for 3D scenarios.

Most of them use a physical-space placement strategy and hence fail to solve viewing problems.

The latest algorithm employs a view-space placement strategy,

However, all of them can not place streamlines that are indeed evenly spaced in the output image.In this paper, we present our interactive view-driven ESS algorithm called IVDESS.

IVDESS is built on our previous ADVESS algorithm.

ADVESS is a 2D engine for sample-based streamline placement.

Here IVDESS is well suited for 3D exploration of PLANAR and SURFACE flows in a perspective-view setting.

IVDESS differs from previous work in that streamlines are really evenly spaced in the output image.

In addition, IVDESS is unique for temporal coherence and high performance.The basic idea of IVDESS is to place streamlines in such a way that they may be non-uniform in physical space.while they are uniform in view space after perspective projection. Here is such a streamline placement of a planar flow that is investigated in 3D.

Surface rendering produces depth information that can be exploited to provide a projection-unprojection pair. Then this pair is used to link physical space and view space.

Streamlines are integrated in physical space while streamline density control is performed in view space.

Specifically, for each newly generated point of a streamline, it is projected to view space to undergo inter-sample distance checking.

The result, either acceptance or refusal, determines whether the streamline is further integrated or immediately terminated.Here is the pipeline of IVDESS, with ADVESS as the underlying engine.This is the hybrid-space multi-level seeding component. View-space seeding can be used to create individual frames while the addition of physical-space seeding maintains inter-frame coherence.

This component performs physical-space streamline integration. A streamline may be integrated from scratch or readily reused from the previous frame. Anyway, each streamline point is projected to view space.

In this component, each line segment is uniformly sampled by the threshold separating distance.Then view-space inter-sample distance control is performed to achieve loop detection and streamline density control.

Streamline integration and inter-sample distance control continue until any sample is rejected. Then the view-space streamline length is checked to determine whether or not the streamline is accepted by the placement and introduces some candidate seeds.

Then the next seed is extracted to continue the placement process until all the seeding queues are empty.Now lets talk about the view-space seeding scheme. It is a temporally incoherent seeding strategy, or TISS.

TISS can be use to generate individual frames by using a double-queue seed scheduler,which means the primary queue takes priority over the secondary queue in providing candidate seeds.

Only when the primary queue is temporarily empty is the secondary queue used to init the placement process or guarantee view coverage.

Candidates introduced by the seed sample of a streamline are stored in the primary queue and are sorted by the view-space streamline length.

Candidates introduced by regular non-seed samples of a streamline are simply appended to the tail of the secondary queue.TISS is an intra-frame view-space seeding mechanism without addressing explorative issues.

Now lets build on top of TISS to maintain smooth inter-frame transition.

In fact, we employ an inter-frame physical-space seeding scheme prior to TISS to make a temporally coherent seeding strategy or TCSS.

The basic idea is to reuse and possibly lengthen the streamlines of the previous frame under normal inter-sample distance control.Specifically, each streamline of the previous frame is processed beginning with the seed in both directions to undergo reprojection, resampling, and possible lengthening.

It is potentially reused in either direction as long as the first in-view segment sample passes inter-sample distance check.

This strategy is greedy in that a streamline even with the seed out of the view may be considered for reuse.

Accordingly, if a streamline passes a view-space streamline length check, the accepted in-view part, the rejected in-view part, and even the out-of-view part are all saved.In fact, the first in-view segment sample is either a raw segment sample when the seed is in the view

or an intermediate segment sample when the seed is out of the view.

Point projection and segment sampling continue until any sample is rejected by inter-sample distance check.A streamline is lengthened when it has been reused through the end of one direction.

Finally the view-space length check is performed to determine whether or not the streamline is accepted by the placement and saved in a buffer.

Non-split streamline reuse+lengthening can prevent the number of streamlines from excessively increasing.

In addition, it suppresses artifacts over the view boundaries and allows closed loops to form in the placement.

In fact, this greedy scheme is very efficient because point projection and the sample-in-view check are computationally cheaper than inter-sample distance check.A successfully reused streamline may include an out-of-view part and a rejected in-view part. However, they should not be rendered to the output image.

So we propose a view-sensitive streamline representation to save physical-space raw points from the negative end to the positive end.

Besides the main body, there is a header in the streamline buffer. The information includes:

In particular, we use two View-Sensitive Descriptors or VSDs, one for each direction, to indicate the accepted in-view parts of a streamline.

VSDs provide a general description of the accepted viewable parts of a streamline to enable greedy reuse and lengthening.

So there may be redundancy between fields and field padding may be needed. And the fields of a VSD need to be dynamically updated to keep track of the change.

What is important is that the use of VSDs can avoid jaggy lines resulting from unprojection errors.

In fact, the unprojection point of a line-view clip sample is temporarily stored in a VSD.

Otherwise the unintended unprojection point might be reused in subsequent frames to cause jaggy lines by lengthening the streamline.

Here we can see some jaggy lines and they were not caused by rasterization.

Currently we have implemented IVDESS for planar flows

and tested placement speed, placement quality, and temporal coherence on a low-end PC.

We used the Northeast Pacific Ocean flow field as the test data. Here is the ocean bathymetry. And here is the complex flow pattern.

Here are the parameters we used for the test.

We produced two movies using IVDESS-TCSS or IVDESS by default and IVDESS-TISS, respectively, based on exactly the same exploration of the flow.This image demonstrates the placement quality of our IVDESS algorithm.

In particular, three closed streamlines were successfully detected and properly formed.This image shows the capability of the TCSS seeding strategy in placing visually uniform streamlines around critical points.This figure shows number of streamlines per frame.

For more than half of the TCSS frames, there are far more reused streamlines than advected ones per frame. Even for the other frames, the number of reused streamlines is only a little bit less than that of advected ones per frame.This demonstrates the effectiveness of the streamline reuse+lengthening scheme of TCSS

The total number of streamlines in a TCSS frame is very similar to that in a TISS frame. This indicates the high-performance of TCSS in preventing the number of streamlines from excessively increasing.

This figure shows streamline reuse percentage for each TCSS frame.

The high percentages demonstrate the effectiveness of the streamline reuse+lengthening scheme adopted in TCSS.

This figure shows frame generation time and frame generation+rendering time.

For nearly every frame and for either case, less time was consumed by TCSS than by TISS.

In addition, the variation in frame generation time for TCSS is much less than that for TISS and this is also the case with frame generation+rendering time.

This figure shows frame generation rate and frame generation+rendering rate.

The interactive and nearly constant frame rates of TCSS indicates that IVDESS-TCSS or IVDESS by default is well suited for coherent flow exploration.Now let me conclude this talk.

IVDESS provides a physically non-uniform but visually uniform representation of planar and surface flows in a perspective-view setting.

A projection-unprojection pair is employed to link physical-space streamline integration and view-space streamline density control.

Greedy but efficient non-split streamline reuse+lengthening is used as a physical-space seeding scheme on top of view-space seeding to make a temporally coherent seeding strategy.

A view-sensitive streamline representation is adopted to support greedy streamline reuse while guaranteeing proper rendering.

IVDESS can achieve coherent flow exploration at interactive frame rates on a low-end PC without pre-processing or GPU support.

As for future work, we plant to enhance IVDESS in support of slows on curvilinear grids and unstructured grids.

In addition, we will investigate the extension of IVDESS for volume flow exploration.