[Insight-developers] Modifying MaskImageFilter to work on VectorImage

[Bradley Lowekamp]

I very much would like to get this filter working with VectorImages.

I don't want this to get forgotten. I think I will be able to take a stab at it based on some of Cory's work in github.

The same technique should be applied to the NotMaskImageFilter as well, are there other similar masking filters?

Brad


On Apr 12, 2011, at 4:12 AM, Gaëtan Lehmann wrote:

> 
> Le 12 avr. 11 à 07:33, Cory Quammen a écrit :
> 
>> Hi Gaëtan,
>> 
>> Thank you for your thoughtful comments.
>> 
>> 2011/4/11 Gaëtan Lehmann <gaetan.lehmann at jouy.inra.fr>:
>>> 
>>> Hi Cory,
>>> 
>>> I think this code would fail if the mask filter is run a second  
>>> time with a
>>> vector image with a different number of components.
>>> There should be a check for the number of components in
>>> BeforeThreadedGenerateData().
>> 
>> Yes, it would be good to check that the number of components in the
>> OutsideValue matches the number in the image, throwing an exception if
>> the number of components doesn't match.
>> 
>>> The initialization of the default value of OutsideValue is quite
>>> problematic, as it can be done only when the number of components  
>>> is known.
>> 
>> Agreed!
>> 
>>> A better option would be to have a way to say "I want to use the  
>>> default
>>> value" - in that case the value can be generated in
>>> BeforeThreadedGenerateData() and would fit the vector the filter is
>>> currently working on - or "I want to use that value" and then it is  
>>> the
>>> responsibility of the user to provide the right number of components.
>>> 
>>> One way to implement that is to use a 3rd input for the filter  
>>> which would
>>> be of type
>>> 
>>> SimpleDataObjectDecorator< PixelType >
>>> 
>>> and to not provide a default value by default - just keep the 3rd  
>>> input as
>>> NULL. If there is no 3rd input, then the default value of  
>>> OutsideValue can
>>> be generated in BeforeThreadedGenerateData(). If there is a 3rd  
>>> input, then
>>> just use the value provided.
>> 
>> While your suggestion sounds reasonable, the same behavior can also be
>> achieved using the current ITK norm of setting filter constants via a
>> method. All it would take would be to examine the OutsideValue to see
>> if it is zero length or if it consists of all zeros, and if so, infer
>> that the user wants to use the default value with the correct number
>> of components. To get back to the "default" value, all the user would
>> have to do is set the OutsideValue to some vector of zeros or a
>> zero-length vector. Does that seem reasonable?
>> 
>>> An additional benefit of that approach: the OutsideValue can be  
>>> provided
>>> through the pipeline, which is not the case currently.
>> 
>> I think that your notion of treating filter constants as inputs to the
>> pipeline is very interesting and has merit, but I'll defer to more
>> experienced developers to determine if that is where ITK should be
>> heading.
>> 
> 
> Hi Cory,
> 
> You're right, this is not the purpose of this change.
> Your proposition to use the default value for a zero length vector  
> seems quite good to me.
> 
> Gaëtan
> 
> 
>> Thanks,
>> Cory
>> 
>>> 
>>> 
>>> Le 11 avr. 11 à 07:30, Cory Quammen a écrit :
>>> 
>>>> David and Luis,
>>>> 
>>>> I have posted a potential solution on github.
>>>> 
>>>> 
>>>> https://github.com/cquammen/ITK/commit/a92bcc79f267d79f48abb7358881c4b2323ae01d
>>>> 
>>>> I will post it to gerrit if you don't see any showstoppers.
>>>> 
>>>> Thanks,
>>>> Cory
>>>> 
>>>> On Sun, Apr 10, 2011 at 5:50 PM, Luis Ibanez <luis.ibanez at kitware.com 
>>>>> 
>>>> wrote:
>>>>> 
>>>>> Hi David,
>>>>> 
>>>>> We have been discussing this problem with Julien Malik from the
>>>>> Orfeo Toolbox team.
>>>>> 
>>>>> In summary, there is a function in the NumericTraits that is  
>>>>> intended
>>>>> to deal with management of size in containers of variable size
>>>>> such as the VariableLengthVector and the std::vector.
>>>>> 
>>>>> Regarding your questions, please see below.
>>>>> 
>>>>> On Thu, Apr 7, 2011 at 7:50 PM, David Doria  
>>>>> <daviddoria at gmail.com> wrote:
>>>>>> 
>>>>>> We have been having some discussion on the best way to allow
>>>>>> MaskImageFilter
>>>>>> to work on VectorImage (with pixels of type  
>>>>>> VariableLengthVector) . The
>>>>>> problem is that the length of a VariableLengthVector is not  
>>>>>> defined at
>>>>>> compile time. This means that in the MaskInput functor, the
>>>>>> member m_OutsideValue cannot be initialized.
>>>>>> 
>>>>>> Brad has identified two problems,
>>>>>> 
>>>>>> 1) How do we initialized the member variable in the functor?
>>>>>> 
>>>>> 
>>>>> We usually add "Set" methods to the functor,
>>>>> the add "Set" methods to the Filter itself,
>>>>> and they simply call the "Set" methods of the functor.
>>>>> 
>>>>>> This seems to be the main issue I'm not sure how to address.
>>>>>> 
>>>>>> 2) When do we detect the size of the vector image, and use  
>>>>>> reasonable
>>>>>> defaults?
>>>>>> 
>>>>> 
>>>>> 
>>>>> At run time,
>>>>> by asking them for their "GetNumberOfComponents()
>>>>> from the vector image.
>>>>> 
>>>>> Reasonable default (otherwise) is "1".
>>>>> 
>>>>>> We think this could be done in BeforeThreadedGenerateData.
>>>>>> 
>>>>> 
>>>>> Yes, that's a good place.
>>>>> 
>>>>> The GenerateInputInformation() or
>>>>> GenerateOutputInformation() may
>>>>> be good places as well.
>>>>> 
>>>>> 
>>>>>> The complete conversation is here:
>>>>>> 
>>>>>> http://review.source.kitware.com/#change,1338
>>>>>> Any thoughts are welcome!
>>>>>> Thanks,
>>>>>> 
>>>>>> David

========================================================
Bradley Lowekamp  
Lockheed Martin Contractor for
Office of High Performance Computing and Communications
National Library of Medicine 
blowekamp at mail.nih.gov


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/mailman/private/insight-developers/attachments/20110422/5f01e680/attachment.htm>