[Rtk-users] I0EstimationProjectionFilter

[Simon Rit]

Hi,
Yes, you need to modify it, add the TOutputImageType to overwrite the
default line 80 of rtkLUTbasedVariableI0RawToAttenuationImageFilter.hxx
<https://github.com/SimonRit/RTK/blob/master/code/rtkLUTbasedVariableI0RawToAttenuationImageFilter.hxx>.
But note that bitshift 2 (the default) seems small to me for unsigned int
because your histogram will be 8 GB. You probably want to test different
bitshifts in this function and use a larger one for int. Actually, you'll
also have a similarly huge lookup table. Have you considered casting your
projections to unsigned short for the I0 estimation?
Simon

On Wed, Mar 8, 2017 at 12:03 PM, Lotte Schyns <lotte.schyns at maastro.nl>
wrote:

> Hello Simon,
>
> Thanks for fixing it. The reason I removed the default templates was
> because there were unsigned shorts in the default templates and for an
> instance of the I0EstimationProjectionFilter class in rtk::
> LUTbasedVariableI0RawToAttenuationImageFilter::BeforeThreadedGenerateData(),
> only TInputImage is defined, so I wanted to prevent some undesired unsigned
> shorts to show up there because of the default templates. Do you think
> LUTbasedVariableI0RawToAttenuationImageFilter can now also handle
> unsigned ints in BeforeThreadedGenerateData, even with the default
> templates? I wasn't sure, so I deleted them.
>
> Lotte
>
> On 07-03-17 22:38, Simon Rit wrote:
>
> Hi Lotte,
> You had removed the default templates, which was the cause of the
> compilation problem. I have fixed it and put the change in a new branch:
> https://github.com/SimonRit/RTK/tree/Lotte1990-patch
> I'll merge tomorrow if the tests pass.
> Thanks for the work,
> Simon
>
> On Mon, Mar 6, 2017 at 1:02 PM, Lotte Schyns <lotte.schyns at maastro.nl>
> wrote:
>
>> Hello Simon,
>>
>> Thanks for the quick response. I made some changes and sent you a pull
>> request. However, because of the changes I made, I'm now having some
>> problems in rtk::LUTbasedVariableI0RawToAttenuationImageFilter::BeforeThreadedGenerateData().
>> I tried to solve it using typename, but it seems to be unsuccessful. Could
>> you please have a look? The I0EstimationProjectionFilter seems to work now.
>>
>> Lotte
>>
>>
>> On 04-03-17 11:25, Simon Rit wrote:
>>
>> Hi Lotte,
>> This code has only be used for unsigned short. I guess you should replace
>> all "unsigned short" in the two code files by InputImagePixelType which is
>> already defined in the header. I would also advise a code review, for
>> example 16 line 33 of rtkI0EstimationProjectionFilter.hxx should be
>> replace by std::numeric_limits
>> <http://en.cppreference.com/w/cpp/types/numeric_limits><InputImage
>> PixelType>::digits.
>> I0EstimationProjectionFilter is based on a histogram analysis. Using 2^16
>> bins for unsigned shorts or 2^32 for unsigned int is too large so bitShift
>> is used to reduce the number of bins to, e.g., for unsigned shorts,
>> 2^(16-bitShift) bins.
>> If you can't figure it out, let us know. If you do, please share your dev!
>> Thanks,
>> Simon
>>
>> On Fri, Mar 3, 2017 at 2:39 PM, Lotte Schyns <lotte.schyns at maastro.nl>
>> wrote:
>>
>>> Hello,
>>>
>>> I would like to set the maximum pixel value in
>>> rtk::I0EstimationProjectionFilter to (2^32)-1 since I have (unsigned)
>>> pixels of 4 bytes each. However, the input for SetMaxPixelValue requires
>>> an unsigned short, so I can only set a maximum value of (2^16)-1, which
>>> is not enough. Is there a possible solution/workaround for this? Could
>>> you also tell me what the bitShift represents in
>>> rtk::I0EstimationProjectionFilter? Thanks in advance.
>>>
>>> Lotte
>>> _______________________________________________
>>> Rtk-users mailing list
>>> Rtk-users at public.kitware.com
>>> http://public.kitware.com/mailman/listinfo/rtk-users
>>>
>>
>>
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/rtk-users/attachments/20170308/caca4613/attachment-0001.html>