Create new 3D bioimaging example. #7309
Conversation
mkcor
added
the
📄 type: Documentation
|
I'm not sure why the second .npy file raises gt = np.load(io.BytesIO(resp.content), allow_pickle=True)but, as expected: Anyway, here's a screenshot of the missing figure: |
| # Ensure TGMM result is an image of type "labeled" | ||
| assert gt.dtype in [np.uint16, np.uint32, np.uint64] | ||
| assert gt.min() == 0 | ||
| assert gt.max() == np.unique(gt).shape[0] - 1 |
There was a problem hiding this comment.
@lagru maybe, once we type the API, the above could turn into a one-liner looking like assert isinstance(gt, Labels)? 💪
There was a problem hiding this comment.
I'm not sure, isinstance() works with class inheritance, typing is not necessarily about that but might be about structural typing and such. And I find it difficult to imagine that running a type checker would check the actual content like it's done here.
There was a problem hiding this comment.
What is structural typing? 👀
There was a problem hiding this comment.
Have a look at protocols and PEP 544 -- Protocols: Structural subtyping (static duck typing) if you are interested. Basically it's trying to address the problem that typing doesn't support a lot of the more dynamic duck typing capabilities of Python. It's also referred to as "static duck typing".
Co-authored-by: ana42742 <amit.ritu.ananya@gmail.com>
| .. [1] McDole K, Guignard L, Amat F, Berger A, Malandain G, Royer LA, | ||
| Turaga SC, Branson K, Keller PJ (2018) "In Toto Imaging and | ||
| Reconstruction of Post-Implantation Mouse Development at the | ||
| Single-Cell Level" Cell, 175(3):859-876.e33. | ||
| ISSN: 0092-8674 | ||
| :DOI:`10.1016/j.cell.2018.09.031` |
There was a problem hiding this comment.
Not sure if that's supported but what do you think about moving this out of the way to the end of the example?
| # | ||
| # data = klb.readfull('Mmu_E1_CAGTAG1.TM000184_timeFused_blending/SPM00_TM000184_CM00_CM01_CHN00.fusedStack.corrected.shifted.klb') | ||
| # sample = data[400:450, 1000:1750, 400:900] | ||
| # np.savez_compressed('sample_3D_frame_184.npz', sample) |
There was a problem hiding this comment.
I was originally looking at the rendered example and was wondering / confused on why this code was passing even though pyklb isn't available when building the docs.
This might confuse readers in a similar manner; is this really relevant here? Or how we can make it more clear that this is not part of the examples workflow?
There was a problem hiding this comment.
The rendering of such code blocks used to make it clear that they weren't part of the current workflow... They would appear as 'commented out,' sort of. I had used this formatting in a previous example ("Estimate anisotropy in a 3D microscopy image"). I can see that it's not the case anymore (with the new theme, I guess).
Yes, it would be best to find some alternative formatting...
There was a problem hiding this comment.
Maybe a code block within a citation block? 🤔
| # The sample dataset is a 3D image with 50 `xy` sections stacked along `z`. Let us | ||
| # visualize it by picking one such section in five. | ||
|
|
||
| data_montage = ski.util.montage(im3d[::5], grid_shape=(2, 5), padding_width=5) |
There was a problem hiding this comment.
Always surprised when discovering another small convenience function. 😄
There was a problem hiding this comment.
Taking a step back, I'm curious about your (teaching) goals with this example.
At first glance, the segmentation approach taken here seems very similar to the one taken in Segment human cells (in mitosis). Is the goal to compare the approach to a machine-learning-based one?
There was a problem hiding this comment.
At first glance, the segmentation approach taken here seems very similar to the one taken in Segment human cells (in mitosis).
Yes, but in 3D. I also had in mind the comparison between this 'native' 3D segmentation and a 2D version along z ('stitching' back the xy sections together afterwards): I tried quickly, and the direct 3D segmentation works much better. Also, I'd like to extend this example one day (or write a new one which would refer to this one) with the tracking challenge, since the (original/full) dataset is not only 3D in space but has a time dimension.
Is the goal to compare the approach to a machine-learning-based one?
Yes!
Co-authored-by: Lars Grüter <lagru+github@mailbox.org>
Description
This tutorial is humbly inspired by this monumental paper [1]; check out the video abstract!
This is WIP so, for now, I've just uploaded data to another repo and I haven't even used permalinks. I haven't written up the narrative parts and section titles (yet). I just wanted to share the data analysis workflow without further ado.
@ana42742 @decorouz @lagru feedback welcome!
[1] Katie McDole, Léo Guignard, Fernando Amat, Andrew Berger, Grégoire Malandain, Loïc A. Royer, Srinivas C. Turaga, Kristin Branson, Philipp J. Keller (2018) "In Toto Imaging and Reconstruction of Post-Implantation Mouse Development at the Single-Cell Level" Cell, 175(3):859-876.e33. ISSN: 0092-8674 https://doi.org/10.1016/j.cell.2018.09.031
Checklist
./doc/examplesfor new featuresRelease note
Summarize the introduced changes in the code block below in one or a few sentences. The
summary will be included in the next release notes automatically: