[Insight-users] Memory problem
Lodron, Gerald
Gerald.Lodron at joanneum.at
Wed Apr 13 10:29:56 EDT 2011
You can use the
itkMemoryProbesCollectorBase
to watch your memory. I also use time probe. Define following as global on top of your code:
itk::MemoryProbesCollectorBase g_oMemorymeter;
itk::TimeProbesCollectorBase g_oChronometer;
#define itkProbesStart( text ) std::cout << "Begin: " << text << std::endl; g_oChronometer.Start( text ); g_oMemorymeter.Start( text )
#define itkProbesStop( text ) std::cout << "End: " << text << std::endl; g_oChronometer.Stop( text ); g_oMemorymeter.Stop( text )
#define itkProbesReport( stream ) g_oChronometer.Report( stream ); g_oMemorymeter.Report( stream )
In your code you can use it like this
itkProbesStart( "Step1")
do something
itkProbesStart( "Step1-1")
do something
itkProbesStop( "Step1-1 )
itkProbesStart( "Step1-2 )
do something
itkProbesStop( "Step1-2 )
itkProbesStop( "Step1 )
itkProbesStart( "Step2")
itkProbesStop( "Step2")
itkProbesReport(std::cout);
________________________________
Von: insight-users-bounces at itk.org [mailto:insight-users-bounces at itk.org] Im Auftrag von Melanie Uks
Gesendet: Mittwoch, 13. April 2011 16:25
An: insight-users at itk.org
Betreff: Re: [Insight-users] Memory problem
Thanks for your replies.
Neil, I splitted the image and repeated the process (see the code at the end of my e-mail). The strange thing I totally don't understand is why the processing is working when I run just this processing in a separate programm and not when I include the processing as a "function" in my acutal program.In the separate program it is just working when I split the image in 4 parts.
Gerald, I used now for every filter: filter->SetReleaseDataFlag(true);
There was no difference... unfortunately :-(
2011/4/13 Lodron, Gerald <Gerald.Lodron at joanneum.at<mailto:Gerald.Lodron at joanneum.at>>
itk uses pipelining which means that there is one image in memory after each filtering. So when you have
->image->filter1->filter2->filter3
you get the memory of the image 4 times. If you want to release the memory after each filter step you can use the release data flag of each filter (SetReleaseDataFlagOn). Then you hage the memory only one time
If you have an image of 300x300x512x3 with float (4 byte) you have 527 MB memory per image, windows 32 theoretically only allows propgramms 2GB (in my tests this theoretically is 1.2 GB in practice) so you only can have the image stored into two vairables, in the third time the allocation will fail!
Another solution is to use the RequestedRegion instead of largestpossibleregion but this is not supported for all filters. here only a part of an image is loaded and processed...I have no experience with this
best regards
________________________________
Von: insight-users-bounces at itk.org<mailto:insight-users-bounces at itk.org> [mailto:insight-users-bounces at itk.org<mailto:insight-users-bounces at itk.org>] Im Auftrag von Melanie Uks
Gesendet: Mittwoch, 13. April 2011 15:34
An: insight-users at itk.org<mailto:insight-users at itk.org>
Betreff: Re: [Insight-users] Memory problem
Sorry, I don't have that much expertice: what does "individual chunks" mean and how to implement?
2011/4/13 Neil Panjwani <paniwani at gmail.com<mailto:paniwani at gmail.com>>
I had similar memory allocation problems when running several CT slices in 32 bit. If you can, using 64-bit should solve your memory problems.
Otherwise, I think you'd have to apply your algorithm on individual chunks at a time and combine them at the end.
On Wed, Apr 13, 2011 at 9:22 AM, Melanie Uks <meluks2010 at googlemail.com<mailto:meluks2010 at googlemail.com>> wrote:
I am running 32 bit.
2011/4/13 Neil Panjwani <paniwani at gmail.com<mailto:paniwani at gmail.com>>
Are you running in 32 or 64 bit?
On Wed, Apr 13, 2011 at 4:47 AM, Melanie Uks <meluks2010 at googlemail.com<mailto:meluks2010 at googlemail.com>> wrote:
Hi all,
I tried to filter a 3D image (300x300x512) with the CannyEdgeFilter. I had the problem that I could not allocate the memory for the filtering. Anyway I don't like the CannyEdgeFilter Implementation as one has to guess which values to take for the thresholding before the gradientmagnitude image exists... Therefore I wanted to write my own edge detection routine.
The steps:
I generate a image (size: 300x300x512)
I generate a vector image (size: 300x300x512, vector dimension:3)
a) I split the image, process the imagepart with GradientRecursiveGaussianImageFilter
b) I paste the processed imagepart in the vector image with the PasteImageFilter
I repeat a) and b) until the whole image is processed
Now the first question: Why can't I process the complete image. I have to split the image and get a result... This is the error I get:
itk::ExceptionObject (0151C728)
Location: "class itk::CovariantVector<float,3> *__thiscall itk::ImportImageConta
iner<unsigned long,class itk::CovariantVector<float,3> >::AllocateElements(unsig
ned long) const"
File: h:\itk\insighttoolkit-3.16.0\code\common\itkImportImageContainer.txx
Line: 188
Description: Failed to allocate memory for image.
Ok, if I accept that I have to split the image, I have a second question: I was able to run my example programm (code at the end of this mail). Then I copied the code into my actual program. It is a programm with several itk functions and Qt GUI. The only difference is that the image is saved as global variable. I was not able to run the processing. I always had the memory error. Why???
Here now the code for the example processing:
edgetest.cxx
#include "conio.h"
#include "itkImageFileWriter.h"
#include "itkImageFileReader.h"
#include "itkRegionOfInterestImageFilter.h"
#include "itkGradientRecursiveGaussianImageFilter.h"
#include "itkGradientToMagnitudeImageFilter.h"
#include "itkPasteImageFilter.h"
#include "itkImage.h"
int main(int argc, char* argv[])
{
// Verify number of parameters in command line
// if( argc < 3 )
// {
// std::cerr << "Usage: " << std::endl;
// std::cerr << argv[0] << " inputImageFile outputVectorImageFile " << std::endl;
// return EXIT_FAILURE;
// }
typedef float PixelType;
typedef float ComponentType;
static const unsigned int Dimension = 3;
typedef itk::Image< PixelType, Dimension > ImageType;
typedef itk::CovariantVector< ComponentType,
Dimension > OutputPixelType;
typedef itk::Image< OutputPixelType, Dimension > OutputImageType;
ImageType::Pointer image = ImageType::New();
ImageType::IndexType start;
for(int i = 0; i < Dimension; i++)
start[i]=0;
ImageType::SizeType size;
size[0] = 300;
size[1] = 300;
size[2] = 512;
ImageType::RegionType region;
region.SetSize(size);
region.SetIndex(start);
ImageType::SpacingType spacing;
spacing[0] = 20;
spacing[1] = 20;
spacing[2] = 4.00493;
image->SetRegions(region);
image->SetSpacing(spacing);
image->Allocate();
std::cout << region << std::endl;
OutputImageType::Pointer vec_image = OutputImageType::New();
OutputImageType::RegionType vecregion;
OutputImageType::SizeType vecsize; //Size
vecsize[0] = (image->GetLargestPossibleRegion().GetSize())[0];
vecsize[1] = (image->GetLargestPossibleRegion().GetSize())[1];
vecsize[2] = (image->GetLargestPossibleRegion().GetSize())[2];
std::cout<<"size0: "<< vecsize[0]<< " size1: "<< vecsize[1] << " size2: " << vecsize[2] <<std::endl;
vecregion.SetSize( vecsize );
OutputImageType::IndexType vecstart; //Start
vecstart[0] = (image->GetOrigin())[0];
vecstart[1] = (image->GetOrigin())[1];
vecstart[2] = (image->GetOrigin())[2];
std::cout<<" start0: "<< vecstart[0]<< " start1: "<< vecstart[1] << " start2: " << vecstart[2] <<std::endl;
vecregion.SetIndex( vecstart );
vec_image->SetRegions(vecregion);
vec_image->SetSpacing(image->GetSpacing());
vec_image->Allocate();
// The image buffer is initialized to a particular value
OutputImageType::PixelType initialValue;
// A vector can initialize all its components to the
// same value by using the Fill() method.
initialValue.Fill( 0.0 );
// Now the image buffer can be initialized with this
// vector value.
vec_image->FillBuffer( initialValue );
std::cout<< "Allocate" << std::endl;
typedef itk::RegionOfInterestImageFilter< ImageType, ImageType > ROIFilterType;
ROIFilterType::Pointer roifilter = ROIFilterType::New();
// Number of Regions
int splitcnt_x = 2;
int splitcnt_y = 2;
int overlap = 15;
int stepcnt_x = (int) (size[0]*1.0/splitcnt_x + 0.5);
int stepcnt_y = (int) (size[1]*1.0/splitcnt_y + 0.5);
ImageType::IndexType roistart;
roistart[2]=0;
ImageType::SizeType roisize;
roisize[2]=512;
ImageType::RegionType roiregion;
for (int cnt_x = 0; cnt_x < splitcnt_x; cnt_x++)
{
for (int cnt_y = 0; cnt_y < splitcnt_y; cnt_y++)
{
roistart[0]= cnt_x*stepcnt_x - overlap;
roistart[1]= cnt_y*stepcnt_y - overlap;
if(cnt_x == 0)
{
roistart[0] = 0;
roisize[0] = stepcnt_x + overlap;
}
else
{
roisize[0] = stepcnt_x + 2*overlap;
}
if(roisize[0]+roistart[0] > size[0])
{
roisize[0] = size[0]-roistart[0];
}
if(cnt_y == 0)
{
roistart[1] = 0;
roisize[1] = stepcnt_y + overlap;
}
else
{
roisize[1] = stepcnt_y + 2*overlap;
}
if(roisize[1]+roistart[1] > size[1])
{
roisize[1] = size[1]-roistart[1];
}
roiregion.SetSize(roisize);
roiregion.SetIndex(roistart);
std::cout << "cnt_x: " << cnt_x << " cnt_y: " << cnt_y << std::endl;
std::cout << roiregion << std::endl;
std::cout << "ROI region inside image region is " << region.IsInside(roiregion) << std::endl;
roifilter->SetRegionOfInterest(roiregion);
roifilter->SetInput(image);
//Filter class is instantiated
typedef itk::GradientRecursiveGaussianImageFilter<ImageType, OutputImageType> GradFilterType;
GradFilterType::Pointer gradfilter = GradFilterType::New();
//sigma is specified in millimeters
gradfilter->SetSigma( 1.5 );
// processing pipeline:
gradfilter->SetInput(roifilter->GetOutput());
typedef itk::PasteImageFilter <OutputImageType, OutputImageType > PasteImageFilterType;
// The SetDestinationIndex() method prescribes where in the first
// input to start pasting data from the second input.
// The SetSourceRegion method prescribes the section of the second
// image to paste into the first.
OutputImageType::IndexType destinationIndex;
destinationIndex[0] = cnt_x*stepcnt_x;
destinationIndex[1] = cnt_y*stepcnt_y;
destinationIndex[2] = 0;
PasteImageFilterType::Pointer pasteFilter = PasteImageFilterType::New ();
pasteFilter->SetSourceImage(gradfilter->GetOutput());
pasteFilter->SetDestinationImage(vec_image);
OutputImageType::RegionType pasteregion;
OutputImageType::IndexType pastestart;
pastestart[2]=0;
OutputImageType::SizeType pastesize;
pastesize[2]=512;
pastestart[0]= overlap;
pastestart[1]= overlap;
pastesize[0] = stepcnt_x;
pastesize[1] = stepcnt_y;
if(cnt_x == 0)
{
pastestart[0] = 0;
}
if(cnt_y == 0)
{
pastestart[1] = 0;
}
pasteregion.SetIndex(pastestart);
pasteregion.SetSize(pastesize);
pasteFilter->SetSourceRegion(pasteregion);
pasteFilter->SetDestinationIndex(destinationIndex);
try
{
pasteFilter->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_SUCCESS;
}
}
typedef itk::GradientToMagnitudeImageFilter< OutputImageType, ImageType > MagFilterType;
MagFilterType::Pointer magfilter = MagFilterType::New();
magfilter->SetInput(vec_image);
try
{
magfilter->Update();
image = magfilter->GetOutput();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_SUCCESS;
}
}
getch();
return EXIT_SUCCESS;
}
_____________________________________
Powered by www.kitware.com<http://www.kitware.com>
Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html
Kitware offers ITK Training Courses, for more information visit:
http://www.kitware.com/products/protraining.html
Please keep messages on-topic and check the ITK FAQ at:
http://www.itk.org/Wiki/ITK_FAQ
Follow this link to subscribe/unsubscribe:
http://www.itk.org/mailman/listinfo/insight-users
_____________________________________
Powered by www.kitware.com<http://www.kitware.com>
Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html
Kitware offers ITK Training Courses, for more information visit:
http://www.kitware.com/products/protraining.html
Please keep messages on-topic and check the ITK FAQ at:
http://www.itk.org/Wiki/ITK_FAQ
Follow this link to subscribe/unsubscribe:
http://www.itk.org/mailman/listinfo/insight-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20110413/30d29d76/attachment-0001.htm>
More information about the Insight-users
mailing list