[Insight-users] Filter re-use problems

[Zachary Pincus]

Karthik,

Thanks for your email!

I tried the mask->Update() bit, and that actually causes an exception. 
Strange.

Here's function 2 again:
Function 2:
   mask = m_CurrentPosition.Value();
   (0)
   [...compute transform...]
   m_Resampler->SetTransform(translate);
   m_Resampler->SetSize(newSize);
   m_Resampler->SetOutputOrigin(newOriginPoint);
   (1)
   m_Resampler->SetInput(mask);
   (2)
   m_Resampler->Update();
   m_CurrentPosition.Value() = m_Resampler->GetOutput();

If I put mask->Update() at position 0 or 2, there is no change. If I 
put it at position 1, an exception gets thrown:
> itk::InvalidRequestedRegionError (0x13017f0)
> Location: "Image::PropagateRequestedRegion()"
> File: 
> /Developer/Local/ITK/ITK-CVS-src/Insight/Code/Common/itkDataObject.cxx
> Line: 397
> Description: Requested region is (at least partially) outside the 
> largest possible region.

I then instrumented my code as requested, at the following locations:

Function 1:
   mask = m_CurrentPosition.Value();
   (0)
   [...compute transform...]
   m_Resampler->SetTransform(affineReverseTransform);
   m_Resampler->SetSize(size);
   m_Resampler->SetOutputOrigin(lowerLeft);
   (1)
   m_Resampler->SetInput(mask);
   (2)
   m_Resampler->Update();
   (3)
   m_CurrentPosition.Value() = m_Resampler->GetOutput();
   (4)

Function 2:
   mask = m_CurrentPosition.Value();
   (5)
   [...compute transform...]
   m_Resampler->SetTransform(translate);
   m_Resampler->SetSize(newSize);
   m_Resampler->SetOutputOrigin(newOriginPoint);
   (6)
   m_Resampler->SetInput(mask);
   (7)
   m_Resampler->Update();
   (8)
   m_CurrentPosition.Value() = m_Resampler->GetOutput();
   (9)


The output is as follows. (Also note that in both [...compute 
transform...] portions of the code, I use a MomentsCalculator object 
with mask as the input to determine the transform, and I use 
mask->TransformPhysicalPointToIndex() and vice-versa. Other than that, 
I don't mess with the mask object.)

0  MaskMT: 92 ResamplerMT: 81
1  MaskMT: 92 ResamplerMT: 122
2  MaskMT: 92 ResamplerMT: 124
3  MaskMT: 92 ResamplerMT: 124
4  MaskMT: 92 ResamplerMT: 124
5  MaskMT: 130 ResamplerMT: 124
6  MaskMT: 130 ResamplerMT: 139
7  MaskMT: 130 ResamplerMT: 140
8  MaskMT: 146 ResamplerMT: 141
9  MaskMT: 146 ResamplerMT: 141

The only odd thing that I notice is that the mask is modified by the 
second call to m_Resampler->Update(), but not the first.

I can send my (slightly nasty) code if that would help...

Thanks,

Zach Pincus

Department of Biochemistry and Program in Biomedical Informatics
Stanford University School of Medicine



On Feb 5, 2005, at 6:01 PM, Karthik Krishnan wrote:

> Hi Zach,
>
> Could you insert the line
>    mask->Update();
> right before the
>      m_Resampler->SetInput(mask);
> line in Function 2
>
> and see if it works.
>
> Could you also please print out
>    std::cout << mask->GetMTime() << std::endl;
>    std::cout << m_Resampler->GetMTime() << std::endl;
> at a bunch of positions, to help debug the problem.
>
> But I think the first solution may work.
>
> Thanks
> kk
>
> Zachary Pincus wrote:
>
>> Hello,
>>
>> I'm trying to re-use a ResampleImageFilter in a tight loop of a 
>> program. Unfortunately, I get corrupted output from the second use of 
>> the same filter each iteration. Only if I make a new resampler each 
>> time can I get valid output.
>>
>> Here's the story: each iteration, I call two functions that use the 
>> same ResamleImageFilter to perform two transforms on an image that's 
>> being cached as m_CurrentPosition.Value(). (Yes, this is terrible 
>> design -- I should be able to compose the transforms, and/or I should 
>> be using a pipeline of two resample filters. However, I've got my 
>> reasons for doing things this way.)
>>
>> Here's what it looks like:
>>
>> Function 1:
>>   mask = m_CurrentPosition.Value(); // fetch the cached mask
>>   [...compute transform...]
>>   m_Resampler->SetTransform(affineReverseTransform);
>>   m_Resampler->SetSize(size);
>>   m_Resampler->SetOutputOrigin(lowerLeft);
>>   m_Resampler->SetInput(mask);
>>   m_Resampler->Update();         m_CurrentPosition.Value() = 
>> m_Resampler->GetOutput(); // cache the newly-transformed mask
>>
>> Function 2:
>>   mask = m_CurrentPosition.Value(); // fetch the cached mask, which 
>> was transformed by function 1
>>   [...compute transform...]
>>   m_Resampler->SetTransform(translate);
>>   m_Resampler->SetSize(newSize);
>>   m_Resampler->SetOutputOrigin(newOriginPoint);
>>   m_Resampler->SetInput(mask);
>>   m_Resampler->Update();
>>   m_CurrentPosition.Value() = m_Resampler->GetOutput(); // cache the 
>> doubly-transformed mask
>>
>>
>> Now, unless I make a new resampler filter before function 2 is 
>> called, I get very corrupted output from function 2. (The image is 
>> wisted and torn, with an incorrect origin.) Note that if I do not 
>> make a new resampler, then in the next iteration (with a new image in 
>> m_CurrentPosition), function 1 again computes good output, and 
>> function 2 again fails. So it's not like the resampler is permanently 
>> broken. It seems to be something I'm doing in function 2.
>>
>> Does anyone have any ideas here? I would be most grateful!
>>
>> Thanks,
>>
>> Zach Pincus
>>
Department of Biochemistry and Program in Biomedical Informatics
Stanford University School of Medicine
>>
>> _______________________________________________
>> Insight-users mailing list
>> Insight-users at itk.org
>> 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
>