Editing labels

[ziw-liu]

New widget contribution to edit labels.

Usage

Prequisite

The dataset must have one channel that contains labels. If this channel is not an integer type, it will be casted to uint16 before viewing and saving. Please report back if your dataset has more than 65535 instances per FOV.

Launch

• From command line: napari -w napari-iohub "Edit labels"
• From napari GUI: 'Plugins' -> 'napari iohub' -> 'Edit labels'

Load FOV

Click 'Browse dataset' and select the dataset with images and labels to edit. Select FOV in the drop-down menus and click 'Load'. This could take a while if loading large images.

Edit

Use napari tools to edit the labels layer.

Save

For each time point, click 'Save' to save the work. This will only save the current time point in the current FOV.

• If the save path is the same as input, this will edit in-place. A warning will be issued. The labels will be casted to the same data type of the original array, even if it is not an integer type.
• If the save path is a new zarr store, the unsaved timepoints and FOVs in the source dataset will not appear in the saved annotations.

[Soorya19Pradeep]

Thank you @ziw-liu for the quick iteration on this tool. I have tried it and it works great for annotation. I created an editable install of the branch in a fresh conda environment. I had to pip install Napari separately. Is the tool supposed to be pip-installed into an environment with Napari in it? Thanks again!

[mattersoflight]

I also just tested the plugin and it works well! The UI is intuitive except for a user having to set the label layer through the environment variable.
Can we add a dropdown widget for Label Layer to Edit that is populated when Load Dataset or Load FOV are clicked?

The UI is responsive when loading a single FOV from the Exp_2023_09_28_DENV_A2.zarr annotation dataset, but slow when loading a single FOV from 2024_02_07_ZIKV_DENV-Live. I expected that the first FOV will be slow to load (parsing of metadata), but subsequent FOVs will be responsive (loading the pixel data from a path). I was comparing the speeds within the same datasets.

I think this is ready to merge after above issues are handled.

Thanks @ziw-liu !

[ziw-liu]

Napari will be installed automatically when doing pip install -e ".[dev]". It is not required for 'user' installations following napari's contribution guidelines.

[ziw-liu]

The UI is responsive when loading a single FOV from the Exp_2023_09_28_DENV_A2.zarr annotation dataset, but slow when loading a single FOV from 2024_02_07_ZIKV_DENV-Live. I expected that the first FOV will be slow to load (parsing of metadata), but subsequent FOVs will be responsive (loading the pixel data from a path). I was comparing the speeds within the same datasets.

Note that we are experiencing major slow downs on the Lustre FS right now. Testing I/O lentency may not yield consistent results.

[mattersoflight]

Good point. Can Soorya and you also test the speed? May be we keep a realistic dataset in login node's scratch space and test the plugin against it before we merge. We can also discuss strategies for automating such tests when we talk today.

…

On Mon, Mar 18, 2024, 9:23 AM Ziwen Liu ***@***.***> wrote: The UI is responsive when loading a single FOV from the Exp_2023_09_28_DENV_A2.zarr annotation dataset, but slow when loading a single FOV from 2024_02_07_ZIKV_DENV-Live. I expected that the first FOV will be slow to load (parsing of metadata), but subsequent FOVs will be responsive (loading the pixel data from a path). I was comparing the speeds within the same datasets. Note that we are experiencing major slow downs on the Lustre FS right now. Testing I/O lentency may not yield consistent results. — Reply to this email directly, view it on GitHub <#3 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAWMLJ3RU7FPQT2XT4LWMN3YY4IJ3AVCNFSM6AAAAABEYVSU3KVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMBUGM3TGNBYGY> . You are receiving this because you commented.Message ID: ***@***.***>

[Soorya19Pradeep]

@mattersoflight , lustre has been very slow since Friday. @biohubgriznog has updated it on #hpc-community slack channel. Maybe we can check with him when it will be back on track so that we can do a better speed test.

[mattersoflight]

@Soorya19Pradeep @ziw-liu loading 2D movies is reasonably fast! I tried the segmentations from a recent experiment /hpc/projects/intracellular_dashboard/viral-sensor/2024_02_07_ZIKV_DENV-Live/4-segment/nuclei_mask.zarr/.

@ziw-liu when the Label Channel was set to Nuclei_Mask, the layer was not cast as labels layer. Can you add that behavior?

@Soorya19Pradeep as we discussed, the correct training/test dataset can be derived from uninfected, high-MOI, and mid-MOI FOVs.

@ziw-liu In the process, I noticed that the nuclei in the above dataset are under-segmented.

Can you and Soorya figure out the movies that capture different infection conditions and are properly segmented?

[ziw-liu]

@mattersoflight Did you click 'Load' to load the image?

napari-iohub.mov

Also the ROI you showed is not under-segmented:

[mattersoflight]

@ziw-liu LGTM. After @Soorya19Pradeep has annotated one movie with this, the widget is ready for other teams to try.

[ziw-liu]

@Soorya19Pradeep Can I have your review?

[Soorya19Pradeep]

I have annotated two movies each with 48 frames using the plugin. It works well for me! Thanks @ziw-liu!