[Insight-users] Failed to allocate memory for image

Mon Oct 2 05:22:48 EDT 2006

Hi Luis,

I tried adding

        metric->SetComputeGradient( false );

before invoking Intialize(), but at run-time, program crashes in file
itkMeanSquaresImageToImageMetric.txx, line 262 with the error:'Image
gradient is null, maybe you forgot to call Initialize()'.

Is it possible to not compute gradient with this registration method?
I tried to change GradientRecursiveGaussianImageFilter with
GradientImageFilter but it needs some changes... I'll try this when I'll
be less busy.

Let me know if there is a simpler way to avoid crash.

Thaks,

Laurent.

> Hi Laurent,
>
>
> Could you please try the following ?
>
> add the call:
>
>
>        metric->SetComputeGradient( false );
>
> before invoking Initialize()
>
> It is likely that the filter that computes
> the image gradients is demanding too much
> memory.
>
> This filter: the GradientRecursiveGaussianImageFilter
> uses an internal minipipeline that in 3D has 2
> smoothing filters and one derivative filter. The total
> process of computing the gradient of your image can
> easily allocate six times the original size of the image,
> but with float pixels. So, if your images were using
> 1byte pixel types, then the gradient filter requires
> 24 times the amount of memory of the input image.
> This would be clearly outside of the capability of your
> RAM (although... your swap should have helped here...).
>
> This filter is greedy in memory, in order to be fast
> in execution. For large images, this may not be a wise
> choice. You may have to replace the filter with the
> GradientImageFilter.
>
>
>
> Please let us know what you find.
>
>
>     Thanks
>
>
>        Luis
>
>
> -------------------
> Laurent PAUL wrote:
>> Hi Ming, hi Luis,
>>
>> Ming seems to have the same problem than mine.
>>
>> I register two volumes ~110Mo each, but program crashes, with the same
>> exception object.
>> I'm sure this registration had ever worked one time with the same
>> computer
>> (even with larger volumes)!
>>
>> I try adding 'metric->Initialize()' inside a try/catch block.
>> First, I added:
>>
>>   metric->SetFixedImage ( FixedImage );
>>   metric->SetFixedImageRegion(FixedImage->GetLargestPossibleRegion());
>>   metric->SetMovingImage( MovingImage );
>>   metric->SetTransform(transform);
>>   metric->SetInterpolator(  interpolator  );
>>
>> because itkMeanSquaresImageToImageMetric was complaining.
>>
>> Then I caught an exception, the same than I had before:
>> ------------------------------------------------------
>> itk::ExceptionObject (0117F854)
>> Location: "unknown"
>> File:
>> C:\lib\ITK\InsightToolkit-2.8.1\Code\Common\itkImportImageContainer.txx
>> Line: 188
>> Description: Failed to allocate memory for image.
>> -------------------------------------------------------
>>
>> Like Ming, I got lot more memory than needed (2Go).
>> I monitored its use with task manager:
>>
>> up to 1100Mo total memory usage ( 800 Mo for registration program) after
>> reading both fixed and moving volumes.
>> It goes down to 770 Mo total usage (470 Mo for registration) after
>> destruction of intermediate filters (casters).
>>
>> At this point, program crashes with error message doing
>> metric->Initialize().
>>
>> I compared MeanSquaresImageToImageMetric and itkImportImageContainer.txx
>> in both distribution, ITK 2.6 and ITK 2.8.1, they are identical except a
>> #include "itkImportImageContainer.h".
>>
>> Is there any changes in code which produce this behavior?
>> Is there any windows limitation about memory usage for one application?
>>
>> Let me know if you want more details.
>>
>> Thanks,
>>
>> Laurent.
>>
>>
>>
>>
>>
>>
>> Hi Ming,
>>
>> Thanks for providing the answers to our questions.
>>
>>  From the additional information that you posted, it
>> seems that the main suspect may be the computation
>> of the gradient image that is performed during the
>> initialization of the ImageMetric.
>>
>>
>> Please add a call:
>>
>>       metric->Initialize()
>>
>> just before you call
>>
>>        registrationMethod->StartRegistration()
>>
>> and put a try/catch() block aroung the metric->Initialize()
>> call, along with a very unique message in the catch block.
>>
>> Then rerun your registration.
>>
>> In this way we should be able to identify if the problem
>> happens during the initialization of Metric.
>>
>>
>> Please let us know what you find.
>>
>>
>>
>>     Thanks
>>
>>
>>
>>       Luis
>>
>>
>> -------------------
>> Ming Chao wrote:
>>
>>>Hi Luis,
>>>
>>>Sorry for my late response. Please see the following for your questions.
>>>
>>>
>>>What tool did you used for monitoring the memory usage ?
>>>
>>>The tool I used is just the Windows Task Manager.
>>>
>>>What was the memory usage for the entire machine ?
>>>(you only reported the usage for the registration application).
>>>
>>>I don't remember the exact number, but I do know that the registration
>>
>> application had the largest consumption of memory.  I am sure the total
>> memory usage did not exceed the allowed memory.
>>
>>>Do you have other applications running at the same time that
>>>your are running the registration application ?
>>>
>>>No.
>>>
>>>In Windows 32 bits, a process can only access 2Gb of contiguous
>>>RAM.
>>>
>>>What Image Metric are you using ?
>>>(e.g. is it one of the metrics that computes the image gradient?)
>>>
>>>Normalized Cross Correlation.
>>>
>>>What other ITK and/or VTK filter do you have in that pipeline.
>>>
>>>I did not have any other filters.
>>>
>>>At what point in the execution of your program do you get this
>>>message ?
>>>
>>>Before the first iteration during the registration which is actually a
>>
>> rigid one.
>>
>>>Have the registration started ?
>>>
>>>Yes.
>>>
>>>Do you got some optimization iterations ?
>>>
>>>No.
>>>
>>>It is very likely that you are actually using most of those
>>>2Gb of data, and may have to release memory as your appllication
>>
>> progresses with its pipeline.
>>
>>>I am not sure what you meant here. Why do you say that I used most of
>>
>> the 2GB data? How can I release memory?
>>
>>>I want to repeat that my registration is actually very simple one, ie,
>>
>> rigid registration. I don't have any intermediate filters. Hope my
>> answers to your questions help to debug the problem.
>>
>>>Thanks,
>>>Ming
>>>
>>>
>>>
>>>--------------------
>>>Ming Chao wrote:
>>> > Hi,
>>> >
>>> > I know this question has been asked many times, but it's still not
>>
>> clear
>>
>>> > to me how to solve it. I try to register two images which are ~70mb
>>
>> each
>>
>>> > (512X512X136). I got the the following error:
>>> >
>>> > ===================================
>>> > ExceptionObject caught !
>>> >
>>> > itk::ExceptionObject (0129FB70)
>>> > Location: "unknown"
>>> > File: C:
>>>\ITK\InsightToolkit-2.8.1\Code\Common\itkImportImageContainer.txx
>>> > Line: 188
>>> > Description: Failed to allocate memory for image.
>>> >
>>> > ===================================
>>> >
>>> > I am using ITK 2.8.1, VTK 5.01, VC++ 6.0. The operating system is
>>
>> Windows XP professional edition (32 bits) with Xeon CPU 3.6GHz and
>> 3.0
>>
>>> > GB memory. I monitored the memory usage and found that only 280 MB
>>
>> used
>>
>>> > for the registration process which is nothing unreasonable. What is
>>
>> wrong here? How to solve the problem. I know if I filtered the image
>> by
>>
>>> > 2X2, then it would be fine, but I would still like to keep the image
>>
>> as
>>
>>> > it is.
>>> >
>>> > Thanks for any hints,
>>> > Ming
>>> >
>>> >
>>> >
>>> > ------------------------------------------------------------------------
>>> >
>>> > _______________________________________________
>>> > Insight-users mailing list
>>> > Insight-users at itk.org <mailto:Insight-users at itk.org>
>>> > http://www.itk.org/mailman/listinfo/insight-users
>>><http://www.itk.org/mailman/listinfo/insight-users>
>>>
>>>
>>
>>
>>
>>
>> _______________________________________________
>> Insight-users mailing list
>> Insight-users at itk.org
>> http://www.itk.org/mailman/listinfo/insight-users
>>
>
>
>