Nrrd ITK-SCIFIO Bug

[Kedar Grama]

Hi Mark,

You are right. I just need access to the multi-series dataset and the
vector image seems like the logical choice. However each series is not a 3D
image. They are 11 2D-images of the same scene at different wavelengths.

Thanks,

-Kedar


On Tue, Jul 2, 2013 at 3:46 PM, Kedar Grama <gbkedar at gmail.com> wrote:

> Hi Mark,
>
> You are right. I just need access to the multi-series dataset and the
> vector image. However each series is not a 3D image. They are 11 2D-images
> of the same scene at different wavelengths.
>
> Thanks,
>
> -Kedar
>
>
> On Tue, Jul 2, 2013 at 3:37 PM, Mark Hiner <hinerm at gmail.com> wrote:
>
>> Curtis - sorry, I actually did the @ encoding to extract specific series
>> earlier today (here<https://github.com/scifio/scifio-itk-bridge/commit/3661b29b2faac4a71f5a625efc882010e84f4a23>)
>> and replied by private e-mail.
>>
>> Kedar can correct me if I'm wrong, but I don't think he explicitly needs
>> Vector images types... he just wants the ability to convert a complete
>> multi-series dataset, and VectorImages seemed to make the most sense to me.
>>
>> You could convert each series in a dataset right now with the @ encoding,
>> but in Kedar's use case (a 65GB .czi file) the entire image gets parsed
>> each time a single series is read and written (of 580 total).. so the time
>> requirements don't seem reasonable. Also, a new file is created for each
>> series (although these could be consolidated with the ImageSeriesReader<http://www.itk.org/Doxygen/html/classitk_1_1ImageSeriesReader.html>I think).
>>
>> But even in Kedar's case where I believe all the 3D images are the same
>> size, the creation of a VectorImage is not simple, because - from what I
>> understand - it expects all the pixel 0 values, then all the pixel 1
>> values, etc... which is basically worst-case-scenario when getting data one
>> series at a time from Bio-Formats (and piped back in arbitrary chunks with
>> no concept of pixel delimitation, at that).
>>
>> If there was an itk::image type that expects series to appear
>> sequentially though, that would at least be a start..
>>
>>
>> On Tue, Jul 2, 2013 at 2:10 PM, Curtis Rueden <ctrueden at wisc.edu> wrote:
>>
>>> Hi Kedar,
>>>
>>> Thanks for exercising the new code! I have a question, and a request.
>>> First, the latter: we just created a new SCIFIO mailing list (CCed). Can we
>>> please discuss these technical issues on the public list from now on?
>>>
>>> And my question: can you describe your need for the Vector image type in
>>> a little more detail? From my (limited) perspective,  using an ITK vector
>>> image type to represent multiple N-D Image blocks, which may have differing
>>> dimensional extents, is fundamentally unworkable. Is there another way to
>>> accomplish your goals?
>>>
>>> Regards,
>>> Curtis
>>>
>>> P.S. I spoke with Mark this morning about a possible alternative
>>> approach for requesting specific Images (i.e. "series") from the SCIFIO
>>> ImageIO: encode it in the file path string somehow.
>>>  On Jul 2, 2013 1:24 PM, "Kedar Grama" <gbkedar at gmail.com> wrote:
>>>
>>>> Hi Mark,
>>>>
>>>> Thanks for working on the vector image support through SCIFIO. I have
>>>> shared a nrrd file with you on google docs. When I read it with the
>>>> ITK-SCIFIO reader it switches the bit ordering LittleEndian/BigEndian
>>>> whereas if I read it in Fiji or build ITK without SCIFIO it is read
>>>> properly. To test this you can use the ITK reader example and write the
>>>> image out as a tiff while building with/without SCIFIO.
>>>>
>>>> Thanks,
>>>>
>>>> -Kedar
>>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://scif.io/pipermail/scifio/attachments/20130702/401995ff/attachment-0001.html>