A Particle System for Interactive Visualization of 3D Flows

1
A Particle System for Interactive
Visualizationof 3D Flows
Authors

• Jens Krüger
• Peter Kipfer
• Polina Kondratieva
• Rüdiger Westermann

Hector M. Garcia
Presented By
2
Problem

• In flow research and industrial practice vector
  field data is one of the key sources for the
  analysis of flow field dynamics
• Visual exploration of complex fields imposes
  significant requirements on the visualization
  system and demands for approaches capable of
  dealing with large amounts of vector valued
  information at interactive rates.
• Previous approaches to virtually explore
  high-resolution flow fields lack the ability to
  simultaneously advect and display large amounts
  of particles.

3
Advection

• Advection is transport in a fluid
• The fluid is described mathematically for such
  processes as a vector field, and the material
  transported is described as a scalar
  concentration of substance, which is present in
  the fluid.
• A good example of advection is the transport of
  pollutants or silt in a river the motion of the
  water carries these impurities downstream.

4
Motivation

• Overcome current methods limitations by
  exploiting features of recent graphics
  accelerators to advect particles in the graphics
  processing unit (GPU).
• Ability to achieve interactive streaming and
  rendering of millions of particles using higher
  order numerical integration schemes.
• Enable the virtual exploration of large fields in
  a way similar to real-world experiments.

5
Motivation
6
Some background info

• GPU Graphics Processing Unit.
• It is a dedicated graphics rendering device.
• GPUs have a highly parallel structure which makes
  them more effective than typical CPUs for a range
  of complex algorithms.

7
More background info

• Recent developments in GPUs include support for
  programmable shaders
• Because most of these computations involve matrix
  and vector operations, engineers and scientists
  have increasingly studied the use of GPUs for
  non-graphical calculations.
• Applications requiring massive vector operations,
  can make use of the massive floating-point
  computational power of a GPU. This can yield
  several orders of magnitude higher performance
  than a conventional CPU.

8
Related Work

• Particle tracing techniques for flow viz have
  been studied intensively.
• Core of these techniques use numerical
  integration schemes.
• In flow viz context, analysis of such schemes
  with respect to stability,accuracy and
  performace.

9
Related Work (contd)

• Particle-based techniques can visualize local
  features in the flow.
• Global imaging techniques for 3D fields can
  illustrate global behavior.
• LIC-methods allow for interactive 2D vector
  fields but no good in 3D flow.

10
Related Work (contd)
11
Methods

• Propose a method for overcoming both computation
  and bandwidth limitations using the GPU.
• Use GPU for advection and rendering computations.
• Use improvements to rendering pipeline.

12
Methods (contd)
13
Methods (contd)

• Using this functionality, particle tracing can be
  performed entirely on the GPU.
• Their method computes intermediate results saves
  them in texture memory and uses them again as
  input to the geometry units to render images in
  the frame buffer.
• Initial particle positions stored in RGB texture
  of size M x N.
• User defines number of particles and appropriate
  texture is generated on the CPU and uploaded on
  the GPU.
• Particle Integration
• Incarnation
• advection

14
Methods (contd)
15
Methods (contd)

• Particle Incarnation
• Transformation
• Birth
• Update
• Particle Advection
• Texture access
• Death test
• Advection
• Reincarnation

16
Methods (contd)

• GPU particle engine for flow viz is implemented
  in Cg

17
Methods (contd)
18
Methods (contd)

• Particle Rendering
• OpenGL SuperBuffer
• Memory object is bound as the current texture
  render target and as a vertex array used to draw
  particle primitives.
• Vertex Texture Fetch
• The key concept is to let the fragment units
  generate textures and to use these textures as
  displacement maps for geometric primitives in
  subsequent rendering passes.

19
Methods (contd)

• Rendering
• Points
• Maximum number of particles stored in video
  memory rendered as color primitives is 250
  million per second
• Oriented Point Sprites
• Used to reveal flow direction.
• Use a sprite texture atlas for arbitrary shaped
  geometry

20
Point Rendering
21
Oriented Point Sprite Rendering
22
Sorting

• Authors implement a GPU sorting network into
  their particle engine.
• Based in the Bitonic merge sort algorithm.
• Well suited for GPU architecture because sequence
  of operations is fixed and not dependent in the
  data to be sorted.

23
What is Bitonic merge sort?

• Data independent sorting method based on the
  bitonic sequence
• A 0-1-sequence is called bitonic, if it contains
  at most two changes between 0 and 1.
• More generally, a sequence of numbers is bitonic
  sequence if it has at most one local maximum or
  one local minimum.
• Examples 1,2,3,4,5 10,6,5,3,1
  3,7,9,8,6,5,4,1 10,8,6,9,12,15,20

24
How does it look like ?
25
Derived Flow attributes

• Velocity
• Divergence
• Enstrophy
• ?2

26
Derived Flow attributes (contd)
27
Visualization Geometry

• Stream Lines
• Ping pong buffer (double buffer)
• Texture samples interpreted as control points
• Draw polylines of T control points
• Stream Ribbons
• Show rotation about the flow axis
• Build a second atlas that contains the other rim
  of each stream line rotating the initial normal
  vector according to the accumulated increment
  angles.

28
Stream Lines
29
Stream Ribbons
30
Evaluation

• Model runs at interactive rates on PC hardware
• It outperforms CPU counterparts
• Show timing statistics to compare their GPU
  implementation vs. CPU.

31
Evaluation (contd)

• Lets take a test drive !

32
Conclusion

• Authors successfully demonstrates advantages of a
  GPU implementation of a particle flow simulation.
• The possibility of integrating numerically and
  data intensive computations for flow analysis
  into the rendering process distinguishes the GPU
  engine from previous approaches.

33
Conclusion (contd)

• Besides particle advection, the engine provides a
  variety of visualization options to visually
  convey relevant structures in 3D steady flow
  fields.
• By using massive particle sets in combination
  with oriented sprites, LIC-like visualizations
  can be achieved at interactive rates. This
  includes higher order integration schemes, thus
  providing numerically accurate particle traces.

34
Questions

• Given the parallel architecture of GPUs would a
  GPU cluster method help for visualizing massive
  global 3D flow visualizations?

35
Questions (contd)

• How would the performance of the visualization
  engine be impacted if the vector field is fed by
  a fully functional numerical model. i.e. ROMS

36
Questions (contd)

• Could their implementation be easily extended to
  non-uniform grids ?