[Insight-developers] Streaming with ImageWriter and IORegion

[Bill Lorensen]

Brad,

Really, we don't know what customers may be using this method. Maybe
even correctly in their own ImageIO class.  Many customers write their
own ImageIO classes. For example, Slicer3 has two of its own ImageIO
classes. To be honest, I have not looked at how it's used in the
ImageFileWriter or current ImageIO classes. I'll take a look over the
next few days.

But, as I said before, we released the code with these methods. Since
we made the mistake, the burden is on us to fix it, not our customers.

Bill

On Fri, Dec 19, 2008 at 5:05 PM, Bradley Lowekamp
<blowekamp at mail.nih.gov> wrote:
> Bill,
> This is a very problematic method; I have strong doubts that anyone is using
> it. It is unclear to me what the appropriate behavior should be. If it's
> current behavior must be maintained then, we should copy the current Write
> method, so that it can be called only when needed if this method has been
> set, Then depreciate the method so that it can be removed later.
> MetaImageIO is the only class that can support this method. Use it with any
> other ImageIO (I think) results in no image, or undefined behavior. I
> believe that this method was a start of implementing streaming that was
> never finished. For it to be used the input to ImageFileWrite must have
> already executed and generated this exact region or else the results would
> be wrong, and undefined. Some of these problems could be reduced if
> ImageFileWrite copies the regionIO to a buffer/cache if needed, then it
> would be up to the ImageIO to do what is asked or barf.
> I would argue that anyone using this method should not be using
> ImageFileWriter and should just use the ImageIO object directly, with it's
> current behavior, but that's not backwardly compatible.
>
> What's be best solution?
>
> Brad
> On Dec 19, 2008, at 3:59 PM, Bill Lorensen wrote:
>
> Brad,
>
> We cannot remove methods from a class that has been released.We should
> do what we can to avoid their improper use.
>
> Bill
>
> On Fri, Dec 19, 2008 at 3:17 PM, Bradley Lowekamp
> <blowekamp at mail.nih.gov> wrote:
>
> Hello,
>
> I have essentially completed these modifications. There is just more testing
>
> to write. All of the current non-IO streaming tests are passing (that should
>
> be). The only ImageIO class that was ready for streamed writing was MetaIO,
>
> so that is currently the only file type that is streaming.
>
> Option number 1, was chosen from bellow. It was the least amount of code,
>
> since it doesn't double the splitters.
>
> On the down side of these changes, I removed
>
> ImageFileWriter::Set/GetIORegion. These were low level methods which are not
>
> compatible with the ITK pipeline, and likely to cause segfaults and crashes
>
> if not used just right.
>
> What is the best way to contribute this code? IJ, attach the diff to Mantis?
>
> Just submit it to CVS, and hope CDash doesn't light up like a xmas tree? I
>
> think it's good quality, just may have missed some nuances of ITK or a
>
> compiler compatibility.
>
> I am excited to be able to process my 13GB RGB data set on my laptop :)
>
> Brad
>
> On Dec 18, 2008, at 12:23 PM, Bradley Lowekamp wrote:
>
> Hello,
>
> I am implementing streaming in ImageFileWriter, hoping that it can get on
>
> the fast track to making it into the repository. I have done most of the
>
> work in ImageFileWriter and now working out some difficulties with
>
> ImageIOBase. According to the design Luis wrote below, IOBase should have a
>
> virtual method to create a "region splitter". The difficulties with this
>
> lies in the differences between ImageIORegion and ImageRegion. Specifically
>
> ImageRegion is templated over dimension for use with Image, where
>
> ImageIORegion is not since ImageIO classes are not. The issues then comes
>
> from with ImageRegionSplitter<#> being templeted. So there becomes two
>
> choices I see:
>
> 1) Use ImageRegionSplitter in ImageIO, this is have the
>
> following declaration:
>
>    template <unsigned int VImageDimension>
>
>    typename ImageRegionSplitter<VImageDimension>::Pointer
>
>    ImageIOBase::GenerateWriteRegionSplitter(unsigned int
>
> numberOfStreamDivisions, const ImageRegion<VImageDimension>
>
> &largestPossibleRegion);
>
> It will also need a virtual helper function and do some dynamic casts, as
>
> well as the ugliness of some macro to create the templated object from a
>
> parameter.
>
> 2) Create a new ImageIORegionSplitter hierarchy, and duplicate the a bunch
>
> of code from ImageRegionSplitter. This will have a smoother interface:
>
> virtual ImageIORegionSpliter::Pointer
>
>        ImageIOBase::GenerateWriteRegionSplitter(unsigned int
>
> numberOfStreamDivisions, const ImageIORegion<VImageDimension>
>
> largestPossibleRegion);
>
> Any thought and suggestions are welcome!
>
> Thanks,
>
> Brad
>
> On Dec 6, 2008, at 6:25 PM, Luis Ibanez wrote:
>
>
> Hi Brad,
>
>
> The functionality of Streaming is not fully implemented
>
> on the ImageFileWriter.
>
>
> The design proposal for implementing this feature is to move
>
> (copy/paste) part of the code that you find in the itkStreamingImageFilter
>
> (in Code/BasicFilters) to the
>
> itkImageFileWriter class.
>
> In particular, the proposal is to use the itkImageRegionSplitter
>
> inside the ImageFileWriter.
>
> During the design discussions, one of the issues that came up
>
> is that only the specific ImageIO classes are qualified to know
>
> what would be the appropriate way to split the data in order to
>
> match the type of blocks that the output image file format can
>
> manage (e.g. DICOM can only stream slices).
>
>
> Therefore, the suggested design is that the ImageFileWriter
>
> will ask the ImageIO class to provide a specific instance of
>
> an itkImageRegionSplitter, and then it will use it in a for
>
> loop for generating sub-regions of the image, and request
>
> them from the preceding image filters.
>
>
> --
>
> Until this is implemented, then your best options, is to do
>
> just what you are doing: using the itkStreamingImageFilter
>
> just before the ImageFileWriter. The drawback of course is
>
> that you have to whole in memory the full-size output image
>
> before being able to write it to disk.
>
>
> Please let us know if you have further questions,
>
>
>   Thanks
>
>
>      Luis
>
>
> -------------------------
>
> Bradley Lowekamp wrote:
>
> Greetings all!
>
> I am trying to stream my data set. I see that MetaIO supports streaming for
>
> reading and writing. But I don't seem to be able to use it for writing. I am
>
> trying to stream my 13GB data set from a ImageFileReader through a
>
> ShrinkImageFilter then write is out with a ImageFileWriter. If I replace the
>
> writer with a StreamingImageFilter it works great, the reader streams and
>
> everything.
>
> To get the writer to stream (or not) I am doing this:
>
> shrinker->UpdateOutputInformation();
>
> RGBVolumeType::RegionType outputRegion =
>
> shrinker->GetOutput()->GetLargestPossibleRegion();
>
> itk::ImageRegionSplitter<3>::Pointer splitter =
>
> itk::ImageRegionSplitter<3>::New();
>
> numberOfSplits = splitter->GetNumberOfSplits(outputRegion, numberOfSplits);
>
> writer->SetFileName( outputFilePN.GetPathName());
>
>  writer->SetInput(shrinker->GetOutput());
>
> for(unsigned int i = 0; i < numberOfSplits; ++i) {
>
>     RGBVolumeType::RegionType streamRegion = splitter->GetSplit(i,
>
> numberOfSplits, outputRegion);
>
>     ioRegion = streamRegion; // sudo code
>
>     writer->SetIORegion(ioRegion);
>
>     writer->Update();
>
>   }
>
> Is this the correct approach? Does anyone have an example of streaming
>
> writing? I think I am going to dig though the ImageFileWriter now to see
>
> what are the updates going on in the pipeline execution. I still a bit fuzzy
>
> on these details, so  may miss something.
>
> Please let me know of any suggestions.
>
> Thanks,
>
> Brad
>
> ========================================================
>
> Bradley Lowekamp  Lockheed Martin Contractor for
>
> Office of High Performance Computing and Communications
>
> National Library of Medicine blowekamp at mail.nih.gov
>
> <mailto:blowekamp at mail.nih.gov>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
>
> Insight-users mailing list
>
> Insight-users at itk.org
>
> http://www.itk.org/mailman/listinfo/insight-users
>
> ========================================================
>
> Bradley Lowekamp
>
> Lockheed Martin Contractor for
>
> Office of High Performance Computing and Communications
>
> National Library of Medicine
>
> blowekamp at mail.nih.gov
>
> _______________________________________________
>
> Insight-developers mailing list
>
> Insight-developers at itk.org
>
> http://www.itk.org/mailman/listinfo/insight-developers
>
> ========================================================
>
> Bradley Lowekamp
>
> Lockheed Martin Contractor for
>
> Office of High Performance Computing and Communications
>
> National Library of Medicine
>
> blowekamp at mail.nih.gov
>
>
> _______________________________________________
>
> Insight-developers mailing list
>
> Insight-developers at itk.org
>
> http://www.itk.org/mailman/listinfo/insight-developers
>
>
>
> ========================================================
>
> Bradley Lowekamp
>
> Lockheed Martin Contractor for
>
> Office of High Performance Computing and Communications
>
> National Library of Medicine
>
> blowekamp at mail.nih.gov
>
>