Difference between revisions of "Documentation/Labs/OpenGLFilters"

From Slicer Wiki
Jump to: navigation, search
 
(10 intermediate revisions by 3 users not shown)
Line 10: Line 10:
 
==  9th June 2016 ==
 
==  9th June 2016 ==
 
Attended by: Davide Punzo, Jean-Christophe Fillion-Robin, Ken Martin, Utkarsh Ayachit and Robert Maynard
 
Attended by: Davide Punzo, Jean-Christophe Fillion-Robin, Ken Martin, Utkarsh Ayachit and Robert Maynard
 +
 +
Based on an original idea of Steve Pieper
  
  
Line 15: Line 17:
 
* no requirement on external library (CUDA or OpenCL).  
 
* no requirement on external library (CUDA or OpenCL).  
 
* performance boosting is fairly good to allow almost interactive filtering up to volume of ~10^8 voxels.
 
* performance boosting is fairly good to allow almost interactive filtering up to volume of ~10^8 voxels.
 +
 +
[[File:2016.06.09 SlicerAstro CPU-vs-GPU Filtering.png]]
 +
<br><small>Source: [https://arxiv.org/abs/1609.03782 Finding faint HI structure in and around galaxies: scraping the barrel (Astronomy and Computing accepted)]</small>
 +
  
  
Line 31: Line 37:
  
 
<pre>
 
<pre>
https://github.com/Punzo/SlicerAstro/tree/master/Smoothing/Logic
+
https://github.com/Punzo/SlicerAstro/tree/master/AstroSmoothing/Logic
https://www.overleaf.com/read/jbxxsqtctpgv
+
https://arxiv.org/abs/1609.03782
 
</pre>
 
</pre>
  
Line 46: Line 52:
 
* handle the I/O communication between the <code>vtkImageData</code> and the GPU Texture using fully the VTK shader infrastructure, since the shader infrastructure is already available in VTK (safe binding, automatic versioning, etc... http://www.vtk.org/Wiki/Shader_In_VTK).
 
* handle the I/O communication between the <code>vtkImageData</code> and the GPU Texture using fully the VTK shader infrastructure, since the shader infrastructure is already available in VTK (safe binding, automatic versioning, etc... http://www.vtk.org/Wiki/Shader_In_VTK).
 
* child classes will handle different approach (e.g. multipass/iterative filtering).
 
* child classes will handle different approach (e.g. multipass/iterative filtering).
 +
 +
Ken Martin provided this structure in VTK: [https://gitlab.kitware.com/ken-martin/vtk/commit/c3a6dfa9023d905151ee61eb83f58bef765d0e2e]
  
 
=== Next steps ===
 
=== Next steps ===
Line 52: Line 60:
 
** Timeline: in 2-3 weeks after his vacation
 
** Timeline: in 2-3 weeks after his vacation
 
** Most likely derived from <code>vtkImageAlgorithm</code>, the class will integrate feature from [https://github.com/Kitware/VTK/blob/master/Rendering/OpenGL/vtkGaussianBlurPass.h VTK/Rendering/OpenGL/vtkGaussianBlurPass.h] to facilitate the use of custom shaders to implement GPU based filters.
 
** Most likely derived from <code>vtkImageAlgorithm</code>, the class will integrate feature from [https://github.com/Kitware/VTK/blob/master/Rendering/OpenGL/vtkGaussianBlurPass.h VTK/Rendering/OpenGL/vtkGaussianBlurPass.h] to facilitate the use of custom shaders to implement GPU based filters.
 +
 +
== Status ==
 +
* Ken wrote the base class for operation filtering on GPU [https://github.com/Kitware/VTK/blob/41fab4e9d7a7b55cd5c7368fee914a74b7bec751/Rendering/OpenGL2/vtkOpenGLImageAlgorithmHelper.h]
 +
* and he gave also an example how to use it: [https://github.com/Kitware/VTK/blob/3530cb6e8a7992ba1c8dbf5652c2b9d0c21bf4cd/Imaging/OpenGL2/vtkOpenGLImageGradient.h]
 +
* An application of the class for filtering can also be found here: [https://github.com/Punzo/SlicerAstro/tree/master/vtkOpenGLFilters]
  
 
= References =
 
= References =

Latest revision as of 08:42, 14 September 2016

Home < Documentation < Labs < OpenGLFilters

Overview

Here we summarize some efforts for enabling image processing (filtering/smoothing) in VTK/Slicer using OpenGL.

The reason behind this page is trying to coordinate the various projects regarding this subject.

Meetings

9th June 2016

Attended by: Davide Punzo, Jean-Christophe Fillion-Robin, Ken Martin, Utkarsh Ayachit and Robert Maynard

Based on an original idea of Steve Pieper


In order to implement efficient filtering, OpenGL solution seems the best that fits our needs:

  • no requirement on external library (CUDA or OpenCL).
  • performance boosting is fairly good to allow almost interactive filtering up to volume of ~10^8 voxels.

2016.06.09 SlicerAstro CPU-vs-GPU Filtering.png
Source: Finding faint HI structure in and around galaxies: scraping the barrel (Astronomy and Computing accepted)



Use Case

Based on the Steve Pieper implementation of the classes:

vtkOpenGLTextureImage
vtkOpenGLShaderComputation

here an Use case:

https://github.com/Punzo/SlicerAstro/tree/master/AstroSmoothing/Logic
https://arxiv.org/abs/1609.03782

Full integration in VTK

For a full integration of the openGLfilters pipeline in VTK, it will be ideal to have a general purpose class:

VTKOpenGLImageAlgorithm

this class will have the duty to:

  • handle the I/O communication between the vtkImageData and the GPU Texture using fully the VTK shader infrastructure, since the shader infrastructure is already available in VTK (safe binding, automatic versioning, etc... http://www.vtk.org/Wiki/Shader_In_VTK).
  • child classes will handle different approach (e.g. multipass/iterative filtering).

Ken Martin provided this structure in VTK: [1]

Next steps

  • Talk with Steve and getting his feedback.
  • Ken will be very happy to contribute a base class vtkOpenGLImageAlgorithm
    • Timeline: in 2-3 weeks after his vacation
    • Most likely derived from vtkImageAlgorithm, the class will integrate feature from VTK/Rendering/OpenGL/vtkGaussianBlurPass.h to facilitate the use of custom shaders to implement GPU based filters.

Status

  • Ken wrote the base class for operation filtering on GPU [2]
  • and he gave also an example how to use it: [3]
  • An application of the class for filtering can also be found here: [4]

References