Fix bugs in visualize segmentation #357

alex-l-kong · 2020-12-01T02:01:59Z

What is the purpose of this PR?

We've received a lot of bugs that have resulted from merging #193 in. This PR encompasses the following 3 issues:

Closes Error in Segment Image Data notebok when making overlay using > 3 total nuclear and membrane channels #284
Closes segmentation_utils.visualize_segmentation error #354
segmentation_utils.visualize_segmentation-file path does not exist #356 is definitely an outlier, I don't think it's worth addressing in this PR

How did you implement your changes

It seems that the bugs aren't completely related to visualize_segmentation but rather on what's being passed in. data_xr instead of data_xr_summed, for example, was incorrectly passed in, which likely caused @awedwards issue.

chan_list verification logic was a little buggy, so that was improved as well.

review-notebook-app · 2020-12-01T02:02:02Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

alex-l-kong · 2020-12-01T03:46:30Z

So this method works to fix visualize_segmentation in the notebook for now (which should address at least @awedwards issue), but the conflicts between channels and compartments is definitely a minor headache. Would be helpful to discuss alternatives for the next code review.

templates/Segment_Image_Data.ipynb

…fovs_to_display to None

ngreenwald

Looks good. Can you update the description with which of the issues this is the closing?

The thing I don't like about visualize_segmentation is that there are two separate things people will want to do.

Cycle through different example images in the notebook, seeing how the overlays look
Save the segmentation output for all FOVs in their dataset.

Right now, they're combined together into the same function. This is partly what lead to the show=true errors. What do you think the best way to split these up would be?

alex-l-kong · 2020-12-01T08:42:48Z

Looks good. Can you update the description with which of the issues this is the closing?

The thing I don't like about visualize_segmentation is that there are two separate things people will want to do.

Cycle through different example images in the notebook, seeing how the overlays look

Save the segmentation output for all FOVs in their dataset.

Right now, they're combined together into the same function. This is partly what lead to the show=true errors. What do you think the best way to split these up would be?

@ngreenwald I think the best way to do this is to have plot_overlay only have imshow calls and visualize_segmentation only have imsave calls. Right now, it's also possible for plot_overlay to save an image and I agree, it's really confusing. We can accomplish this by simply returning the result of what would normally be saved by plot_overlay, then allow visualize_segmentation to save it for us.

This is great for making the functionality clearer, but in terms of the issues people have been bringing up, I don't think it's actually show=True that's causing the problem anymore, but rather other errors (ex. data_xr and not data_xr_summed, quite possibly user-specific data issues and filesystem errors, etc.) that just happen to be seen by this particular call to plot_overlay first. I'm doing more investigating into these to see what the root problems are, but I do believe something is getting propagated into the call to plot_overlay with show=True (especially for #354).

ngreenwald · 2020-12-01T15:42:49Z

The get_topmost_subplot error happens when you create two different plots in the same plotting window. This is happening because of the for loop, where each FOV is getting imshowed. The error goes away if only a single FOV is generated.

Okay great, I agree. I think as a further simplification, we don't even need visualize_segmentation to call plot_overlay for generating the RGB overlay. We can just have a separate cell that directly calls plot_overlay to generate the overlay people care about (the channel overlay).

Once they're done with that, then we can have them call visualize_segmentation (which we'll call something else). That function will loop through, and save the output files for all of the fovs in the xarray

alex-l-kong · 2020-12-01T18:55:17Z

@ngreenwald a few questions:

The call to plot_overlay will display only the result for a single fov at a time, as opposed to multiple at a time?
With this new method, looks like we're only going to be saving _segmentation_borders.tiff and _segmentation_labels.tiff files?
Have we already tried using plt.imshow instead of io.imshow? I think plt.imshow could allow us to display multiple channel overlays at once, although that would be a problem if the user passed in 100 fovs.

ngreenwald · 2020-12-01T18:59:47Z

I think we should move all of the stuff that's in plot_overlay which doesn't involve plotting a multi-color overlay and move it to the visualize_segmentation function, which we can rename to save_segmentation_output or something like that.

Then there will be one cell where the user calls plot_overlay directly, with a single FOV specified, to look at the output. And a separate cell which loops over all of the data and produces the borders.

ngreenwald · 2020-12-01T19:02:38Z

Even better, we should modifiy plot_overlay to be create_overlay. Then it can return the image to displayed in the notebook. save_segmentation_output can then also call create_overlay, but without displaying the image each time.

ngreenwald

Looks good. Just a few last remaining skeletons to remove from the plot function that can be moved into the segmentation_labels function, so that all the plot function is responsible for is taking an image and creating an RGB overlay

ark/utils/plot_utils.py

ark/utils/segmentation_utils.py

ngreenwald

Nice, almost there

ark/utils/plot_utils.py

ark/utils/segmentation_utils.py

alex-l-kong · 2020-12-02T22:53:09Z

@ngreenwald OK, the new channel structure has been added, ready for review.

ngreenwald

Close, just need to update the colors of the input data

ark/utils/plot_utils.py

ark/utils/segmentation_utils_test.py

ngreenwald

Nice stuff

ngreenwald · 2020-12-03T19:24:28Z

ark/utils/plot_utils.py

+            raise ValueError("max 3 channels of overlay supported, got {}".
+                             format(plotting_tif.shape))
+
+        # set first n channels (in reverse order) of formatted_tif to plotting_tif


* Fix bug in Segment_Image_Data: change data_xr to data_xr_summed * Handle cases for both compartments and channels * Make visualize_segmentation easier to understand * Change compartments to channels and add comment allowing user to set fovs_to_display to None * Logic and testing added for new plot overlay workflow * Last commit before nuking previous implementation * Nuke previous implementation * Documentation fix * Address code review * Fix call in notebook * Fix documentation for extract_delimited_names * Address code review comments * Assign to RGB channels in BGR order * Add notebook visualization * Address (hopefully) final review comments

Fix bug in Segment_Image_Data: change data_xr to data_xr_summed

84d3d7b

alex-l-kong self-assigned this Dec 1, 2020

Handle cases for both compartments and channels

a28cc8d

alex-l-kong requested review from ngreenwald and ackagel December 1, 2020 03:46

ngreenwald reviewed Dec 1, 2020

View reviewed changes

templates/Segment_Image_Data.ipynb Show resolved Hide resolved

templates/Segment_Image_Data.ipynb Show resolved Hide resolved

alex-l-kong added 2 commits November 30, 2020 21:37

Make visualize_segmentation easier to understand

5e41e80

Change compartments to channels and add comment allowing user to set …

a744589

…fovs_to_display to None

alex-l-kong requested a review from ngreenwald December 1, 2020 05:50

ngreenwald requested changes Dec 1, 2020

View reviewed changes

Merge branch 'master' into plot_overlay_fix

66e9079

alex-l-kong added 4 commits December 1, 2020 12:07

Logic and testing added for new plot overlay workflow

db0c62c

Last commit before nuking previous implementation

1baac1e

Nuke previous implementation

5fe7acb

Documentation fix

47ac376

alex-l-kong requested a review from ngreenwald December 1, 2020 20:50

ngreenwald requested changes Dec 1, 2020

View reviewed changes

alex-l-kong added 2 commits December 2, 2020 00:16

Address code review

282ea9f

Fix call in notebook

5916842

alex-l-kong requested a review from ngreenwald December 2, 2020 17:56

Fix documentation for extract_delimited_names

b635070

ngreenwald requested changes Dec 2, 2020

View reviewed changes

alex-l-kong requested a review from ngreenwald December 2, 2020 19:17

alex-l-kong added 2 commits December 2, 2020 11:18

Address code review comments

60e75cc

Assign to RGB channels in BGR order

892d165

Add notebook visualization

2c15ca9

alex-l-kong requested review from ngreenwald and removed request for ngreenwald December 2, 2020 22:56

ngreenwald requested changes Dec 3, 2020

View reviewed changes

ark/utils/plot_utils.py Outdated Show resolved Hide resolved

ark/utils/segmentation_utils_test.py Outdated Show resolved Hide resolved

Address (hopefully) final review comments

6b5b1be

alex-l-kong requested a review from ngreenwald December 3, 2020 16:13

ngreenwald approved these changes Dec 3, 2020

View reviewed changes

ngreenwald merged commit dc2c7b8 into master Dec 3, 2020

ngreenwald deleted the plot_overlay_fix branch December 3, 2020 19:26

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix bugs in visualize segmentation #357

Fix bugs in visualize segmentation #357

alex-l-kong commented Dec 1, 2020 •

edited

Loading

review-notebook-app bot commented Dec 1, 2020

alex-l-kong commented Dec 1, 2020 •

edited

Loading

ngreenwald left a comment

alex-l-kong commented Dec 1, 2020

ngreenwald commented Dec 1, 2020

alex-l-kong commented Dec 1, 2020

ngreenwald commented Dec 1, 2020

ngreenwald commented Dec 1, 2020

ngreenwald left a comment

ngreenwald left a comment

alex-l-kong commented Dec 2, 2020

ngreenwald left a comment

ngreenwald left a comment

ngreenwald Dec 3, 2020

Fix bugs in visualize segmentation #357

Fix bugs in visualize segmentation #357

Conversation

alex-l-kong commented Dec 1, 2020 • edited Loading

review-notebook-app bot commented Dec 1, 2020

alex-l-kong commented Dec 1, 2020 • edited Loading

ngreenwald left a comment

Choose a reason for hiding this comment

alex-l-kong commented Dec 1, 2020

ngreenwald commented Dec 1, 2020

alex-l-kong commented Dec 1, 2020

ngreenwald commented Dec 1, 2020

ngreenwald commented Dec 1, 2020

ngreenwald left a comment

Choose a reason for hiding this comment

ngreenwald left a comment

Choose a reason for hiding this comment

alex-l-kong commented Dec 2, 2020

ngreenwald left a comment

Choose a reason for hiding this comment

ngreenwald left a comment

Choose a reason for hiding this comment

ngreenwald Dec 3, 2020

Choose a reason for hiding this comment

alex-l-kong commented Dec 1, 2020 •

edited

Loading

alex-l-kong commented Dec 1, 2020 •

edited

Loading