[OpenSPIM] Multiview registration - no inliers

Mon Aug 29 09:29:03 CDT 2016

Hi Neil and Stephan

Thanks for the tips and all the help! I tried again with a higher
concentration of the 1µm beads, got rid of the bright big beads and
increased the z acquisition distance to increase the region of overlap and
was finally successful!

I hope to meet many of you OpenSPIMmers in Sheffield this week!

Best wishes
Tania

On 25 August 2016 at 15:08, Anthony, Neil <nantho2 at emory.edu> wrote:

> Hi Tania,
>
> I'd agree with Stephan on the larger z stacks. I usually take >300 um, but
> it kinda depends on the field of view of the camera, as at some point
> you're taking images in an area that will no longer overlap.
>
> 400 interest points would be sufficient if they all overlapped, but more
> are required in general. I believe Stephan's articles suggest 1000-2000 to
> get the best results. All depends on overlap percentage.
>
> When defining the beads I aim to use a sigma and threshold that picks out
> the peak of the bead if possible. Stephan, can I ask if it's detrimental
> when the threshold is too low and I'm identifying the same bead on many z
> planes? Also, I assume the sigma value determines the steepness of the
> point, but it still seems to pick up on larger aggregated beads etc, so I'm
> not too sure about that part.
>
> Neil
>
>
>
>
> ------------------------------
> *From:* Stephan.Preibisch at mdc-berlin.de
> *Sent:* Aug 25, 2016 3:41 AM
> *To:* Tania Mendonca
> *Cc:* Anthony, Neil; openspim at openspim.org
>
> *Subject:* Re: [OpenSPIM] Multiview registration - no inliers
>
> Hi Tania,
>
> from looking at the data I agree with Anthony, there are some problems
> which make it hard. But trying different calibrations and not getting a
> result makes me believe that there is maybe no overlap in between the
> stacks. You have quite a few of these big blobs apparently, so is it
> possible that these are different blobs?
>
> If you run it again, make the stacks way bigger in z to make sure that
> there is overlap, increase the laser power, and maybe put some simple auto
> fluorescent sample in there so it is possible to judge if the calibration
> is right and if there is overlap.
>
> Good luck,
> Stephan
> ---
>
> Dr. Stephan Preibisch
> Group Leader
>
> Berlin Institute of Medical Systems Biology of the MDC
> Building 89, 1.08b
>
> email: stephan.preibisch at mdc-berlin.de
> web: http://www.preibisch.net/ <http://fly.mpi-cbg.de/preibisch>
>
> On 24 Aug 2016, at 19:49, Tania Mendonca <tvmendonca1 at sheffield.ac.uk>
> wrote:
>
> Hi Neil
>
> Thanks for your email! The stacks should have the voxel size correctly set
> to 0.235 x 0.235 x 0.750. I'm assuming you've looked at the data on
> µManager as I'm unfamiliar with the ViewSetup tag that you mention. My
> system is not quite an OpenSPIM - I use LabView to control and don't have
> the Picard 4D stage (I use a Piezo for scanning instead).
>
> Your manual transform looks about right, the stacks are about 120º apart
> with the big bead somewhere near the middle. I have tried a few different
> µbeads and the ones in these stacks are 1µm green beads but as you say they
> are nowhere near as bright as the big beads. Perhaps I do need to dump the
> large beads and find something else to try and reconstruct. I've been
> finding about 400-ish interest points in my stacks- is there a magic number
> that works for you? I've been imaging to about 100µm depth, how much would
> you recommend? Don't the stacks start getting too large to handle?
>
> Thanks for the tip on the camera noise! I think I'll try again tomorrow
> with brighter beads, no big beads and larger volumes - I'll let you know
> how it goes!
>
> Best wishes
> Tania
>
> On 24 August 2016 at 17:55, Anthony, Neil <nantho2 at emory.edu> wrote:
>
>> Hi Tania,
>>
>>
>> Can I please ask what the z voxel size is?  I see in the test.xml that
>> you have a voxel size of 0.235 in xy and z for id 0,1,2 in the ViewSetup
>> tag.  When I create a xml from scratch here, setting xy to 0.235 and z to
>> 1.5 (the smallest on the standard Picard 4D stage) I see each id in the xml
>> has a size of “0.235 0.235 1.5”.
>>
>>
>>
>> Stephan, I assume these are the voxel sizes used for the BDV.  Would I be
>> correct in assuming some more and saying that if only one value is seen
>> it’s used for xy and z?
>>
>>
>>
>> …  interlude …  etc. etc.  lalalalala …  coffee etc.
>>
>>
>>
>> I now see what the problem is.  Well, what some of the problems might be.
>>
>>
>>
>> I did a manual transform around the y axis and manually overlapped the
>> big bead.  See attached.  Is this how is should look?  If I have it laid
>> out correctly I think you have a combination of not enough beads, combined
>> with not enough overlap of physical volumes given the low number of beads.
>> On top of that I feel the small bead signal is so low that it’s approaching
>> the camera noise*.
>>
>>
>>
>> It is difficult as the big beads are so bright, and you might want to
>> find a better sample to test with.  I’ve done some testing on spheroids
>> that are really bright, and I had to swap my 100nm tetra spec beads for
>> some really bright 1um green beads to be able to get good signal on the
>> beads and not saturate the sample of interest.
>>
>>
>>
>> I started with just little beads to figure out the basics, and I take
>> just sub-res bead data to get my PSF output from the decon too.
>>
>>
>>
>> Neil
>>
>>
>>
>>
>>
>> * inspect the data stacks with orthogonal view and you’ll see the
>> vertical lines running down the stack.  For this kind of noise you can run
>> a few line scans across your beads and noise to get an idea of the
>> intensity difference and then use remove outliers
>>
>>
>>
>>
>>
>>
>>
>> *From:* OpenSPIM [mailto:openspim-bounces at openspim.org] *On Behalf Of *Tania
>> Mendonca
>> *Sent:* Friday, August 19, 2016 3:31 PM
>> *To:* Stephan.Preibisch at mdc-berlin.de
>> *Cc:* openspim at openspim.org
>> *Subject:* Re: [OpenSPIM] Multiview registration - no inliers
>>
>>
>>
>> Hi
>>
>>
>>
>> Just tried it but it still doesn't work.
>>
>>
>>
>> Thanks
>>
>> Tania
>>
>>
>>
>> On 19 August 2016 at 20:13, Stephan.Preibisch at mdc-berlin.de <
>> Stephan.Preibisch at mdc-berlin.de> wrote:
>>
>> Hi, I just think that in the beginning when you define your dataset the
>> pixel sizes were wrong which potentially corrupts the downstream workflow.
>> Try to start from the very beginning, and define a new dataset and manually
>> put in the correct calibration. This might just work ...
>>
>>
>>
>> All the best,
>>
>> Stephan
>>
>> Sent from my iPhone
>>
>>
>> On Aug 19, 2016, at 21:00, Tania Mendonca <tvmendonca1 at sheffield.ac.uk>
>> wrote:
>>
>> Hi
>>
>>
>>
>> I've just checked to make sure but yes I have corrected the voxel size on
>> the image stacks and there shouldn't be a mistake in that. However, your
>> reply gave me an idea and I did a quick test. I duplicated one of the
>> images and ran the multiview reconstruction plugin for the two duplicates.
>> It worked, so it makes me think that there's nothing wrong with the files
>> themselves and you might be right about the issue being with my voxel
>> depth. I've referred to SVI's Nyquist calculator to workout my step size,
>> is there a recommended step size that people use?
>>
>>
>>
>> Best wishes
>>
>> Tania
>>
>>
>>
>> On 19 August 2016 at 16:58, Stephan.Preibisch at mdc-berlin.de <
>> Stephan.Preibisch at mdc-berlin.de> wrote:
>>
>> Hi, my first guess would be that your calibration (pixel size is x,y and
>> z) is wrong ... Can you make sure this is correct when you define your
>> dataset?
>>
>>
>>
>> All the best,
>>
>> Stephan
>>
>>
>> On Aug 19, 2016, at 17:17, Tania Mendonca <tvmendonca1 at sheffield.ac.uk>
>> wrote:
>>
>> Hi all
>>
>>
>>
>> Hope your microscopes are aligned and treating you well! I'm sorry if
>> I've missed this conversation before but I was wondering if any of you have
>> had similar problems as me with trying to get the Fiji multiview
>> reconstruction plugin to like my data.
>>
>>
>>
>> I've tried a range of microspheres - different colours, different sizes
>> and different concentrations but it never seems to find enough
>> correspondences between views. I've been primarily experimenting with
>> interactively detecting interest points. I've tried the presets as well
>> but with no luck. I've tried to systematically work through various
>> combinations of sigma and threshold but always end up with 0 inliers.
>> Registrations always terminates with the following error: "There are no
>> connected tiles, cannot do an optimization. Quitting."
>>
>>
>>
>> My home built system doesn't run on µManager and my images are acquired
>> as .cxd files which I convert to .tifs before processing. I'm attaching
>> some of my test data -https://drive.google.com/
>> file/d/0B-he-_wcys_-aUtRZ0pidFhyUlk/view?usp=sharing
>>
>> the sample here is 1µm yellow-green beads (1:4000 dilution) and 10µm
>> beads (1:50 dilution) with the idea of using the 1µm beads to reconstruct
>> the 10µm beads. I use the same objectives as on the OpenSPIM system (10X
>> for illumination and 20X for detection), magnification on the system is
>> 27.7X. The illumination here is 473nm and detection at 520nm.
>>
>>
>>
>> A log file has been included in the zip file to give you an idea of what
>> I've been trying.
>>
>>
>>
>> I hope I'm just missing a trick somewhere. Any help would be greatly
>> appreciated!
>>
>>
>>
>> Best wishes
>>
>> Tania
>>
>> --
>>
>> -------------------------
>>
>> Tania Mendonca
>>
>> Doctoral Research Student
>>
>> University of Sheffield
>>
>> S10 2TN
>>
>> -------------------------
>>
>> _______________________________________________
>> OpenSPIM mailing list
>> OpenSPIM at openspim.org
>> http://openspim.org/mailman/listinfo/openspim
>>
>>
>>
>>
>>
>> --
>>
>> -------------------------
>>
>> Tania Mendonca
>>
>> Doctoral Research Student
>>
>> University of Sheffield
>>
>> S10 2TN
>>
>> -------------------------
>>
>>
>>
>>
>>
>> --
>>
>> -------------------------
>>
>> Tania Mendonca
>>
>> Doctoral Research Student
>>
>> University of Sheffield
>>
>> S10 2TN
>>
>> -------------------------
>>
>> ------------------------------
>>
>> This e-mail message (including any attachments) is for the sole use of
>> the intended recipient(s) and may contain confidential and privileged
>> information. If the reader of this message is not the intended
>> recipient, you are hereby notified that any dissemination, distribution
>> or copying of this message (including any attachments) is strictly
>> prohibited.
>>
>> If you have received this message in error, please contact
>> the sender by reply e-mail message and destroy all copies of the
>> original message (including attachments).
>>
>
>
>
> --
> -------------------------
> Tania Mendonca
> Doctoral Research Student
> University of Sheffield
> S10 2TN
> -------------------------
>
>
>

-- 
-------------------------
Tania Mendonca
Doctoral Research Student
University of Sheffield
S10 2TN
-------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openspim.org/pipermail/openspim/attachments/20160829/5b724657/attachment-0001.html>