Thanks @mkcor!

Hello @mkcor! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-04-21 11:24:37 UTC

Now you understand why it's called confocal fluorescence microscopy!! w00t @GenevieveBuckley ❤️

Achieved with:

import plotly.express as px

px.imshow(data[n_plane // 2], zmin=v_min, zmax=v_max)

@emmanuelle @nicolaskruchten thanks for making imshow so powerful and flexible in Plotly Express!! I was looking at PR plotly/plotly.py#2691 and I simply used https://plotly.com/python/imshow/#defining-the-data-range-covered-by-the-color-range-with-zmin-and-zmax (great documentation, btw).

I cannot achieve it with:

from skimage import io

io.imshow(data[n_plane // 2], plugin='matplotlib', vmin=v_min, vmax=v_max)

which roughly amounts to:

import matplotlib.pyplot as plt

plt.imshow(data[n_plane // 2], vmin=v_min, vmax=v_max)

It's as if I didn't pass any min/max values...

@mkcor yeah matplotlib's imshow uses vmin/vmax only for scalar (single-channel) data, it is ignored for multichannel data. We might checks with specialists (@GenevieveBuckley) if it's ok to change the contrast for each channel but since channels are not blended too much intermediate hues are not important. If it's confirmed that it's ok to change the contrast for fluorescence data then I'm really happy that we included this possibility in plotly's imshow. Maybe it's worth opening an issue in matplotlib's tracker to discuss this point?

We might check with specialists (@GenevieveBuckley) if it's ok to change the contrast for each channel but since channels are not blended too much intermediate hues are not important. If it's confirmed that it's ok to change the contrast for fluorescence data then I'm really happy that we included this possibility in plotly's imshow.

@emmanuelle yes, I would also prefer double-checking with a specialist of the field if and how we should change the contrast in a fluorescence microscopy image. Notably, should we do it on a per channel basis? I understand that my doing

px.imshow(slice, zmin=v_min, zmax=v_max)

is equivalent to

px.imshow(
    slice,
    zmin=[v_min, v_min, v_min],
    zmax=[v_max, v_max, v_max]
)

and, while the maximum value happens to be the same across channels (in this case, 4095), the minimum value is not (it goes like [10, 68, 35]). Going for

px.imshow(
    slice,
    zmin=[data[:, :, :, 0].min(), data[:, :, :, 1].min(), data[:, :, :, 2].min()],
    zmax=[v_max, v_max, v_max]
)

doesn't feel great, but maybe it's done...? I would say that

px.imshow(
    slice,
    zmin=[0, 0, 0],
    zmax=[v_max, v_max, v_max]
)

looks most natural but, again, I'm not familiar enough with confocal fluorescence microscopy.

Let's poke some biology experts who might know better!
@GMFranceschini @slowkow @fran6w @ThomasWalter @kmader @IvoLeist

@mkcor yeah matplotlib's imshow uses vmin/vmax only for scalar (single-channel) data, it is ignored for multichannel data.

Oh, right, the docs read "When using scalar data ..." indeed, but I didn't get that the parameter would just be ignored for multichannel data.

Maybe it's worth opening an issue in matplotlib's tracker to discuss this point?

Ok, taking care of it now.

Changing the contrast independently for each channel for visualization is fine.

Great demo! I would also say it's fine to normalize on a per-channel basis and for most cases, it will prevent really poorly scaled images (why is my figure all black/white? questions) and having things just work out of the box is nice.

I might have a note of it since many scikit-image tutorials end up as figures in papers (and comparing two groups with normalized channels might lead to wrong conclusions).

@GenevieveBuckley @kmader thank you for your replies, they are really helpful!

Great demo! I would also say it's fine to normalize on a per-channel basis and for most cases, it will prevent really poorly scaled images (why is my figure all black/white? questions) and having things just work out of the box is nice.

@kmader ok, in this case I will favour the following option:

px.imshow(
    slice,
    zmin=[data[:, :, :, 0].min(), data[:, :, :, 1].min(), data[:, :, :, 2].min()],
    zmax=[data[:, :, :, 0].max(), data[:, :, :, 1].max(), data[:, :, :, 2].max()]
)

I might have a note of it since many scikit-image tutorials end up as figures in papers (and comparing two groups with normalized channels might lead to wrong conclusions).

Oh, I would be very interested in this note! Thank you so much, again.

@mkcor Thank you for the poke! Python, bioinformatics... but I am not a biology expert ;-)

@mkcor glad you thought about me! Unfortunately, I have never worked with images so I can't help with this issue!

@fran6w @GMFranceschini I know your expertise is in bioinformatics rather than cell imaging... I reached out just in case and we got answers anyway, so it's all good! Thanks anyway.

Re-posting:

@scikit-image/core please review. I'm happy to add an entry to the release notes, but I can't figure out which file under doc/release/ I'm supposed to edit... 🤔

I'm happy to add an entry to the release notes, but I can't figure out which file under doc/release/ I'm supposed to edit

I think just doc/release/release_dev.rst.

I think just doc/release/release_dev.rst.

@grlee77 oh, thanks! 84d65a8 I replaced the entry mentioning a previous biology example (#4648) with this current one, because #4648 made it into the 0.18 release so... I guess it should be pruned from release_dev.rst, once it lands in a given release? Not sure because a bunch of other Documentation entries can be found in both release_dev.rst and release_0.18.rst... 😕

@mkcor I think what's happened is that release_dev wasn't pruned after release. That would be my fault. Perhaps look at the log to see what happened in that file, if anything, since 0.18, and delete everything that was there at the 0.18 branch point. (663e9e4, I think.)

Hi @grlee77,

My only question is about how we can make this and other Plotly-based examples more useful for users that try to run them from the command line.

Thank you for keeping this in mind! I admit I forgot about this.

One possibility is calling plotly.io.show on each figure, for example.:

fig = px.imshow(data[n_plane // 2], zmax=vmax)
plotly.io.imshow(fig)

or simply (sparing import plotly):

fig = px.imshow(data[n_plane // 2], zmax=vmax)
fig.show()

If I do that everywhere in the example, I end up with one browser tab per plotly figure.

Right. I think it's quite nice, but...

I didn't check if that will interfere with the gallery example, though? Let me know what you think.

It does! 😭 I tried and ended up with the same display issue as in: https://scikit-image.org/docs/dev/auto_examples/data/plot_3d.html It used to work, so I'm not sure when the issue arose (maybe a particular mix of Plotly and Sphinx versions...?). I'm noticing a display issue with the thumbnail as well.

Adding a plt.show() at the end of the script can also keep the Matplotlib figure from disappearing immediately once the script completes. We have that in many of the existing Matplotlib-based examples.

Good point. In this gallery example, only one figure (i.e., the first one) is a Matplotlib figure... But it's better than nothing: done 1a0a7af.

@mkcor I think what's happened is that release_dev wasn't pruned after release. That would be my fault. Perhaps look at the log to see what happened in that file, if anything, since 0.18, and delete everything that was there at the 0.18 branch point. (663e9e4, I think.)

@jni ok, thanks. The overall cleanup could/should be a different PR, so I have pruned only the Documentation section (37b2b07, a0b0de7) by comparing release_dev.rst and release_0.18.rst side by side. I'll take care of item "Documentation has been added to the contributing notes about how to submit a gallery example" separately (either deleting it or finding corresponding PRs).

PR should be good to go now!

I opened #5341 about pruning the release_dev.rst file so we don't forget

I tried and ended up with the same display issue as in: https://scikit-image.org/docs/dev/auto_examples/data/plot_3d.html

Is there an issue for that? Do you want to open a PR removing the call to show for now so things show up in the gallery as expected?

I tried and ended up with the same display issue as in: https://scikit-image.org/docs/dev/auto_examples/data/plot_3d.html

Is there an issue for that? Do you want to open a PR removing the call to show for now so things show up in the gallery as expected?

Yes, I'll take care of it tomorrow (there's currently no issue for that).