Let's see...
Right now, you do:
RealMaskRealInterval myMask = ...;
RandomAccessible<T> myImage = ...;
IterableInterval<T> myMaskedImage = Regions.sample(myMask, myImage);
Cursor<T> c = myMaskedImage.localizingCursor();
long[] pos = new long[c.numDimensions()];
while (c.hasNext()) {
T value = c.next();
c.localize(pos);
// and now we do something with the position and sample value
}
However, there are some limitations:
myMaskhas to be aRealMaskRealInterval, not aRealMask. This is unfortunate. We could add aRegions.sample(RealMask, RandomAccessibleInterval)that creates aRealMaskRealIntervalmatching the bounds of the given RAI. I think?- This
myMaskedImagecannot be used withLoopBuilderbecauseLoopBuilderonly supports RAI images, not II images. So we have to use this clunkier cursor code, rather than a nice lambda.¯\_(ツ)_/¯
Do you agree with (1) that we could have a better utility method here?
Because I think the most common thing is that you have a
RealMask (not necessarily a RealMaskRealInterval) and a RAI and you want to combine them.
Hm, so far I've only seen examples of this in discrete space, i.e. going via
RandomAccessible and IntervalView, see imagej-roi-course
It makes sense as its own project during the learnathons... but then why not follow up by migrating shared useful materials out of learnathon space and into the official tutorials repo?
Unfortunately, they have been a hodge-podge of git repositories. Which ends up being fine during the event, and really tough afterward to see what's been presented.
@NicoKiaru I have the following workflow:
- I open a XML/H5 (that contains only a single
Source) and view it in BDV. - In BDV I change its
AffineTransformusing the manual transform UI of BDV - Now I would like to save a new
*.xmlfile that looks exactly as the original one that I opened, but I want a new<affine>...</affine>content, reflecting the additional transform that I manually added in BDV.
Do you know how to do this most efficiently? Do I have to construct a new SpimData object from the TransformedSource and then use XmlIoSpimData.save()? If so, would you have code already to construct SpimData from a single Source and the path to the corresponding .h5 file? Or are there other ways to save such an .xml file?
Hi Christian, you can use this
spimData.getViewRegistrations().getViewRegistration(timePoint,iViewSetup).preconcatenateTransform(myAffineTransform); or spimData.getViewRegistrations().getViewRegistration(timePoint,iViewSetup).concatenateTransform(myAffineTransform);, with myAffineTransform being the transform retrieved from the TransformedSource. Then you can resave the dataset. FYI, I also found a trick to update the spimData when it's shown in the bdv (thanks to reflection to access a private method see https://github.com/BIOP/bigdataviewer_scijava/blob/master/src/main/java/ch/epfl/biop/bdv/scijava/command/spimdata/UpdateSpimDataDisplay.java)
For the moment yes, you're correct. If the
loadTimepoint method in AbstractSpimSource was made public, then it would be a bit easier. (https://github.com/bigdataviewer/bigdataviewer-core/blob/5be52618026c6734911d9c150a9e2ba1140799e0/src/main/java/bdv/AbstractSpimSource.java#L187)
to update the final model
@tischi
after concatenating or preconcatenating
it just adds it to the list of transformations
we do not automatically re-compute the final model every time you add
@StephanPreibisch somehow I do not manage. It is not updating the image in BDV. Probably I am doing something wrong:
final Source< ? > source = bdv.getBdvHandle().getViewerPanel().getState().getSources().get( currentSource ).getSpimSource();
final SpimData spimData = sourceToSpimData.get( source );
spimData.getViewRegistrations().getViewRegistration( 0, 0 ).concatenateTransform( new ViewTransformAffine( "My transform", myTransfrom ) );
spimData.getViewRegistrations().getViewRegistration( 0, 0 ).updateModel();
bdv.getBdvHandle().getViewerPanel().requestRepaint();sourceToSpimData is a Map that I generate while adding new SpimData to BDV in order to keep track of things.
@tischi I'm pretty sure you need to do the trick I showed you for the Bdv update when using BdvFunctions (at least for the versions I'm using in the POM file). I don't know how the update is done in BigStitcher. But that would be interesting to know! I guess calling the
updateModelfunction is useful for updating properly the SpimData object before saving it, but I'm nor sure..
if you would close and open BDV it should update
check this code out
Thanks @StephanPreibisch ! And you're also using reflection: you call the non public
SpimSource.loadTimepoints method through reflection
Thanks @NicoKiaru & @StephanPreibisch !
It looks like that both of your codes you are calling a method of the Source that is wrapped by the TransformedSource, which is the object BDV is using to display the source. What I did initially was simply:
final TransformedSource transformedSource = ( TransformedSource ) source;
transformedSource.setFixedTransform( myTransform );
bdv.getBdvHandle().getViewerPanel().requestRepaint();This updates the display of the TransformedSource in the BDV and is much simpler code. I think the difference is that the TransformedSource is something that only exists in BDV and is not talking back to the SpimData object that was initially put into BDV for display. My feeling is that if one just wants to add additional transforms to Sources and display those in BDV, using the TransformedSource.setFixedTransform() approach might be cleaner?! In fact, the only reason I now also want to modify the SpimData object is because I want to save a new xml with myTransform appearing as additional <affine>...<\affine> tags. What do you think?
...what I need for our users is a solution that works consistently for both adding a
ManualTransformation (by pressing T in BDV) and/or adding any kind of other transformation, e.g. as obtained by a phase-correlation algorithm. And they should play well together. E.g. a user might do a rough manual pre-alignment (using T) and then press, e.g., P to compute a fine phase-correlation alignment on top of the manual pre-alignment. I am a bit worried right now that the current ManualTransformation in bdv-core relies on the TransformedSource, while our solutions seem to rather work through SpimData, and thus things might become messy, but I might be wrong as I am still not fully grasping everything...
That's also what I understood @tischi. If you want to resave, I think it's better to go for the
SpimData option. Extra advantage of using SpimData is that the volatile view is taken care of also. It may not be the case fot the TransformedSource ? An alternative : every time you want to save the SpimData, check the transform of the TransformedSource: if it's not identity, then push it to SpimData.
But you may end up with a long chain of transform if the user is doing many attempts.
I am confused :-) I applied the
ViewRegistration in the SpimData and also applied your reflection trick, in above example applying a simple translation along the x-axis. The result in BDV is that the yellow box is shifted along the x-axis relative to the grey square (see screenshot). However the coordinates of the upper left corner of the image are still at x=0 and y=0 if you hover there with the mouse. If I would change the transformation via the TransformedSource.setFixTransform() this would be different: the coordinates of the image would actually be shifted. Can someone explain me the meaning of the gray square and why the coordinates do not change using the SpimData approach? (...for me, I think the coordinates really would need to change because I need to move around the relative location of multiple Sources in a physical coordinate system).
Is there a one-liner in ImgLib2 for copying an image's contents into another image, without using a lambda? I know about imglib/imglib2/pull/237 but that isn't merged. I know about
LoopBuilder but that requires a lambda. I want to do this from Python and I'm concerned that writing the lambda will not be elegant and/or performant. My current approach is to use Ops (ij.op().run("copy.rai", result, rai)) but @haesleinhuepf says in imagej/pyimagej#46 that it is too slow and must be fixed immediately.
@NicoKiaru and @StephanPreibisch ...the solution is that I have to call the loadTimepoint method for both the
Source< ? > and the Source< ? extends <Volatile< ? >>. An alternative to twice calling the private loadTimepoint() method via reflection seems to be to simply use transformedSource.setFixedTransform( transform ). This also properly updates the BDV. I don't understand yet what the pros and cons are...