Polygon issue

[joshmoore]

Dearest @ome/ngff,

My notes-fu is failing me. Several times in the past couple of days someone has asked about the vector/polygon specification for ROIs. However, I only see #33 "mesh specification". @glyg (rightly) threw us off the mesh issue, but I can't remember if someone had already volunteered to startup the new polygon specification issue?

Any takers?
~Josh

NB: chose to try out the GH discussions which tend to be the issue writers/commenters rather than ping all 70+ interested users from image.sc. Let me know if that's a problem.

[glyg]

Hi all,
I can also take on ROIs — but before I start implementing something, we have to decide between GeoJSON and WKT for the underlying spec. Also if someone can have a look at the mesh implementation to see if it at least make sense, this could provide me with some guidance for ROIs

Cheers

Guillaume

[joshmoore]

I don't think I know enough to really respond on #33. But it might make sense to have some samples, either in the issue or up somewhere on s3, so folks can give the different options a whirl.

[joshmoore]

Adding here for when the time comes: https://arxiv.org/abs/2005.06469

[glyg]

This should be fun to implement :)

[glyg]

Some pointers regarding the implementation:

• Shapely is I think the best candidate as an interface & reference
• There's a protocol that should at least be implemented in the reader (if shapely dependence is too much)

[ivirshup]

One downside of shapely: I believe it largely focusses on 2d, with very limited support for 3d.

Oh whoops, just noticed that this is quite an old comment.

[joshmoore]

@glyg: any recent thoughts on this? What would be the best link to give someone who was interested in "multiscale polygons" within the NGFF format?

[glyg]

Hi @joshmoore I haven't thought about that much more but I have been talking with Philippe Roudot @proudot who's interested in storing ROIs maybe he'll have more ideas! Also we wanted to work together on that - particularly storing dynamical ROIs (I'll let him explain in more details).

As for the spec, I feel GeoJSON makes more sense?

[glyg]

Maybe relevant here : (h/t Philippe also):

https://ccmsc.utah.edu/images/publications/presentation-abstracts/publications/Christensen_1_2014_EfficientIOAdaptive.pdf

[proudot]

Hi there,

Sorry, I just saw that notification while I was starting to setup github to manage my code. Better late than never I guess ?

My specific use-case when it comes to ROI is to provide a ROI with arbitrary shape that changes dynamically over time (a.k.a. dynROI). They are equipped with an frame of refence that changes overtime along with the moving polygons to facilitate visualization as well as coordinate mapping. That is implemented in Matlab at https://github.com/DanuserLab/u-track3D and used in various paper on 3D+time data analysis.

That boat may have sailed but let me know if my input can help in any way.

Cheers,
Philippe

[joshmoore]

Sorry, I missed this, @proudot. Looks like the conversation is still goin on over on #33.