How to find contour offset?

[unimusic]

Hi, I am trying to extract information using the xml file to compute the inverse transform of a contour and overlay it with the original image in a a section. I am not trying to use the GUI for RECONSTRUCT, as I would like to do this automatically in python.

My problem is that the x and y coordinates in the image space of a section do not match with the x and y coordinates in the contour space in the same section. I've included example images to try to illustrate the problem.

You can see that in the first image, the coordinates are in pixels and they are of a different number than in the second image, where the coordinates are in microns. I am talking about the 'position' as seen in the bottom bar. The mouse is in the same position for both images.

Where in the data (xml, code, ??) can I find the translation and mapping between these two coordinate systems? Thank you.

[k34harris]

Note -- that one is the domain -- that is in pixels. The other is showing the position of the trace -- which is in microns. those are different components of the same trace file xml for that section Can you send the trace file for section 7661? Kristen *K <http://synapseweb.clm.utexas.edu>risten Harris* *915 Wild Basin Ledge* *Austin, TX 78746* *kmh2249@gmail.com <kmh1149@gmail.com>* *Personal Account*

…

On Wed, Jun 24, 2020 at 12:59 PM Angela Zhang ***@***.***> wrote: Hi, I am trying to extract information using the xml file to compute the inverse transform of a contour and overlay it with the original image in a a section. I am not trying to use the GUI for RECONSTRUCT, as I would like to do this automatically in python. My problem is that the x and y coordinates in the image space of a section do not match with the x and y coordinates in the contour space in the same section. I've included example images to try to illustrate the problem. [image: image] <https://user-images.githubusercontent.com/10647159/85608091-98a2fa00-b609-11ea-9fbe-b078dd5d70ab.png> [image: image (1)] <https://user-images.githubusercontent.com/10647159/85608106-9b055400-b609-11ea-8dfd-f6b7ac800ced.png> You can see that in the first image, the coordinates are in pixels and they are of a different number than in the second image, where the coordinates are in microns. Where in the data (xml, code, ??) can I find the translation and mapping between these two coordinate systems? Thank you. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#4>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADHOUZVSSUIIE4CXKNN2OSLRYI5HRANCNFSM4OHAEP6Q> .

[k34harris]

Also, the coordinate system in Reconstruct is not very usual -- Tom bartol at the Salk might be able to tell you about how the coordinates differ from usual. Namely, I believe 0 is not at a corner, and +/- x, y are not side/side up/down the same as one might expect. I didn't write the code, so I am not 100% sure, but Tom and I figured it out a couple years ago...

[k34harris]

If you do use the Reconstruct Gui, you can export all of your images at once with the traces on them here:

[k34harris]

To do so, you would enter series-> export >images
Then fill in the blanks on the window that opens --
You can export the entire series or just a subset.
You can set the colors, and export the images in any way that you want.
Might be more efficient than Python code to do the same thing.

[k34harris]

Note that you can also have it match the current zoom, or export the whole image.

[unimusic]

It seems that 0,0 in the image space (pixels) is actually 0,0 in the original image, and the axis move along the original space as well. Here is the trace file for series 7661.

Series1.7661.7661.zip

Here is the pre-and post- transformed points in the contour overlaid with the original image:

You can see that neither of these contours match.

We do not want to export anything using the GUI, because we ultimately want to train a deep learning algorithm to do detection of contours, so we need to keep the image and contour information separate. We just want to match them up automatically.

Please advise. Thank you!

[unimusic]

We also tried applying the image transform to the image and the contour transform to the contour, and overlaying them. We scaled the contour (since it was in microns) by 0.002, because that was the only ratio number we could find using the GUI. As you can see, neither the scale nor the position aligns properly.

[k34harris]

I am copying Tom Bartol and Uri Manor, you are reaching the limits of my knowledge. Best wishes, Kristen *K <http://synapseweb.clm.utexas.edu>risten Harris* *915 Wild Basin Ledge* *Austin, TX 78746* *kmh2249@gmail.com <kmh1149@gmail.com>* *Personal Account*

…

On Wed, Jun 24, 2020 at 3:30 PM Angela Zhang ***@***.***> wrote: We also tried applying the image transform to the image and the contour transform to the contour, and overlaying them. We scaled the contour (since it was in microns) by 0.002, because that was the only ratio number we could find using the GUI. As you can see, neither the scale nor the position aligns properly. [image: download (12)] <https://user-images.githubusercontent.com/10647159/85624505-dfe7b580-b61e-11ea-8cad-74f69b9240a1.png> — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADHOUZUKJGZ2NFYUP4ZIUDDRYJO7TANCNFSM4OHAEP6Q> .

[unimusic]

Thank you! Tom has agreed to meet with me next week. It seems that the transforms are polynomial basis transforms and not affine transforms, and they are the 'inverse transform from the image space to the screen space', although I'm not sure what that means. The code is here but it is rather hard to understand due to the variable names not being self explanatory and lack of comments: https://github.com/SynapseWeb/Reconstruct/blob/18adc8af7feb28acafd31046eac5034f4d903106/source/nform.cpp

[k34harris]

I had ment to copy Edwin on this information... He is our local 'new' engineer who will be updating Reconstruct... Best, Kristen *K <http://synapseweb.clm.utexas.edu>risten Harris* *915 Wild Basin Ledge* *Austin, TX 78746* *kmh2249@gmail.com <kmh1149@gmail.com>* *Personal Account*

…

On Wed, Jun 24, 2020 at 6:28 PM Angela Zhang ***@***.***> wrote: Thank you! Tom has agreed to meet with me next week. It seems that the transforms are polynomial basis transforms and not affine transforms, and they are the 'inverse transform from the image space to the screen space', although I'm not sure what that means. The code is here but it is rather hard to understand due to the variable names not being self explanatory and lack of comments: https://github.com/SynapseWeb/Reconstruct/blob/18adc8af7feb28acafd31046eac5034f4d903106/source/nform.cpp — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADHOUZTYKTBSJSFBN375JTDRYKD2HANCNFSM4OHAEP6Q> .