[ome-users] Mirror before stitching?

[Björn Quast]

Hi Sam,


when I look at your attachment, I am pretty sure, that the top row should be 
on the bottom of the stitched image, the second row from top should be the 
second from bottom and so on. So there is no error in your metadata, nor in 
the plugin, its just the interpretation of the 0,0 coordinate. ImageJ should 
interprete 0,0 on left top of an image, your software might use left bottom. 
So no flipped Y-axis in images, just different order of tiles. Trie to fix it by 
telling the plugin to use row by row, left & up.


Best Björn



> Attached is a screenshot of the issue I'm seeing, even when I deselect
> "compute overlap" and set the slider to 0. On the right is the LOCI
> stitching, which seems reasonable except the flipped y axis. The left is
> the output of the plugin, in which images overlap erroneously (?). When I
> "ignore calibration," there is no change.
> 
> Is my metadata wrong? Does the plugin handle composite (multi-channel)
> images well?
> 
> -Sam
> 
> On Jun 19, 2013, at 11:29 AM, Curtis Rueden <ctrueden at wisc.edu> wrote:
> > Hi Sam,
> > 
> > > Also, what does the "ignore calibration" option do?
> > 
> > I would test with this option both on and off. It controls whether the
> > stage position coordinates are interpreted as pixels, or in calibrated
> > physical units (e.g., microns).
> > 
> > Regards,
> > Curtis
> > 
> > 
> > On Wed, Jun 19, 2013 at 1:27 PM, Sam Lord <sjlord at berkeley.edu> wrote:
> > Thanks for the response, Curtis.
> > 
> > I tried setting the overlap slider to 0 and deselecting all other options.
> > I thought this would reproduce the results that the LOCI importer
> > stitching gave me. Instead, the final fused image was about half the
> > dimensions, and several tiles were fused on top of each other. Maybe
> > there's an issue reading the positions from the metadata?
> > 
> > I'll keep playing around with it.
> > 
> > Also, what does the "ignore calibration" option do?
> > 
> > Thanks,
> > 
> > -Sam
> > 
> > On Jun 19, 2013, at 9:39 AM, Curtis Rueden <ctrueden at wisc.edu> wrote:
> >> Hi Sam,
> >> 
> >> > I've been having a lot of trouble getting the Fiji Grid/Collection
> >> > Stitching plugin to operate how I want.
> >> 
> >> Thanks for the report. I am forwarding this thread to fiji-devel, which
> >> is a better place to report issues with Fiji plugins such as
> >> Grid/Collection Stitching.>> 
> >> > -If I deselect "compute overlap," I would expect the plugin to simply
> >> > place the tiles where the stage position in the metadata claims it is.
> >> > Instead, the output seems to not include most of the images.
> >> 
> >> Try reducing the overlap slider to 0. IIRC, it defaults to 10, and that
> >> factor still applies even if you uncheck "compute overlap.">> 
> >> > -When I do select "compute overlap," the images often aren't placed
> >> > anywhere near the right locations. For instance, it will output an
> >> > irregular shape, when the fused image should be a rectangle.
> >> 
> >> There are various ways of tuning how it approaches the computation, but
> >> if your stage coordinates are already very accurate, you probably don't
> >> need the "compute overlap" feature anyway.>> 
> >> > I'm sure some of these issues result from a non-perfect pixel
> >> > calibration and the fact that I did not include any overlap in my
> >> > tiles when acquiring them.
> >> 
> >> Ah, yes. If you do not include any overlap, then the "Compute overlap"
> >> feature will always fail. It relies on overlap between tiles to detect
> >> how things should be laid out. Your only option when collecting fully
> >> non-overlapping tiles is to uncheck "Compute overlap" and reduce the
> >> overlap slider to 0. It should then lay them out exactly how you
> >> collected them.>> 
> >> > However, the LOCI importer seems to not worry about that. Instead, it
> >> > happy tiles my images. The only problem is that I can tell the images
> >> > are flipped in the y direction.
> >> 
> >> Right, the LOCI importer will stitch your image *exactly* how the
> >> metadata says, and cannot do anything else such as flip tiles inverted
> >> in Y. So I suggest you try the Grid/Collection Stitching plugin again as
> >> I described above. If you still cannot get it work, and are willing to
> >> post a sample non-working dataset, we can investigate further.
> >> 
> >> Regards,
> >> Curtis
> >> 
> >> 
> >> On Wed, Jun 19, 2013 at 11:29 AM, Sam Lord <sjlord at berkeley.edu> wrote:
> >> Hi Curtis,
> >> 
> >> I've been having a lot of trouble getting the Fiji Grid/Collection
> >> Stitching plugin to operate how I want. I'm asking it to select
> >> positions from file and instructing it to use the image metadata. Here
> >> are some issues I've found:
> >> 
> >> -The color scheme does not import. Neither does some other metadata (e.g.
> >> pixel size). -If I deselect "compute overlap," I would expect the plugin
> >> to simply place the tiles where the stage position in the metadata
> >> claims it is. Instead, the output seems to not include most of the
> >> images. -When I do select "compute overlap," the images often aren't
> >> placed anywhere near the right locations. For instance, it will output
> >> an irregular shape, when the fused image should be a rectangle.
> >> 
> >> I'm sure some of these issues result from a non-perfect pixel calibration
> >> and the fact that I did not include any overlap in my tiles when
> >> acquiring them. However, the LOCI importer seems to not worry about
> >> that. Instead, it happy tiles my images. The only problem is that I can
> >> tell the images are flipped in the y direction.
> >> 
> >> Is there a way to "brainlessly" tile my images from metadata location AND
> >> flip y?
> >> 
> >> -Sam
> >> 
> >> On Jun 16, 2013, at 11:49 AM, Curtis Rueden <ctrueden at wisc.edu> wrote:
> >>> Hi Sam,
> >>> 
> >>> We recently added an "invert Y axis coordinates" option (and one for X
> >>> as well) to the Stitching plugins. Did you try that?
> >>> 
> >>> The Fiji Grid/Collection Stitching plugin, in case you haven't tried it,
> >>> uses Bio-Formats and is much more powerful than the BF Importer's basic
> >>> stitching feature.
> >>> 
> >>> Regards,
> >>> Curtis
> >>> On Jun 16, 2013 12:01 PM, "Sam Lord" <sjlord at berkeley.edu> wrote:
> >>> Importing stacks from Micro-Manager works well using Bio-Formats
> >>> Importer, and the stacks option seems to read location directly from
> >>> metadata. The only issue I have is that my camera must be rotated 180
> >>> degrees, because the columns stitch together well, but each row is
> >>> upside-down.
> >>> 
> >>> I tried importing the stack, flipping, exporting, then reimorting, but
> >>> something must have gotten lost in the metadata in that maneuver.
> >>> 
> >>> Any chance there is a simple way to tell the stitching that the images
> >>> are flipped?
> >>> 
> >>> -Sam
> >>> 
> >>> _______________________________________________
> >>> ome-users mailing list
> >>> ome-users at lists.openmicroscopy.org.uk
> >>> http://lists.openmicroscopy.org.uk/mailman/listinfo/ome-users