frontend/node-red-dashboard: Allow adjusting the minimum size threshold for segmented objects

[gromain]

Right now, there is no way to change the minimal size of the objects found by the segmenter.

There should be a slider somewhere to set this.

[gromain]

Partially solved for now, uses min mesh size * 2.
No dynamic selection though.

[ethanjli]

I'm adding this feature request to the backlog milestone because @fabienlombard requested it today in a discussion about the segmenter. This feature request will need some thoughtful design about how to represent any default or non-default size threshold in the metadata exported by the segmenter, because this is an important experimental parameter which determines the composition of the data produced by the segmenter (just like mesh filter size determines the composition of objects imaged by the PlanktoScope).

[ethanjli]

This feature request was discussed again in detail in the 2024-10-10 software meeting. @tpollina has made a tentative name for the metadata field to represent the size threshold (though this might be changed): process_min_ESD. He also added an input field for this in one of the settings pages of the new/upcoming (v3) Node-RED dashboard.

@fabienlombard might be the person to make sure we have discussions in future meetings to make design decisions for this feature's UI/UX, e.g. whether users should set the value in units of pixels or um (Fabien says the value should be pixels), and what the the metadata field's units should be; and other design decisions which we haven't thought about yet.

I have volunteered to implement the necessary changes in the existing (v2) Node-RED dashboard if someone else makes the UI/UX design decisions. @fabienlombard might have someone who can work on implementing the necessary changes in the Python segmenter. No timeline has been set for these changes - we will make the changes when we find time to do so, but reminders to keep thinking about this may help us make time.

[ethanjli]

As part of a DM conversation with me on Slack, Fabien says the minimum size threshold should instead be calculated as follows (note that this is a fix for #114):

minESD = minMesh    # or process_min_ESD in future  (in micron here)
        pixel_size = self.__global_metadata.get(“process_pixel”, None) # make the program crash if process_pixel is missing, don't default to 1.0
       minradius_pix = minESD /2 /pixel_size  # now we work in pixels
 minArea = math.pi * (minradius_pix ) * (minradius_pix )

And the threshold should be applied as follows:

and line 487 should be modified from
region for region in regionprops if region.bbox_area >= minsizepix
to
region for region in regionprops if prop.filled_area >= minArea

Here is some follow-up conversation:

Fabien:

normally with this you do the right job
what was done before: incorrectly multiplying by two the size to ignore (compensated by the fact that he was inspecting the area of the full vignette)
(grossly compensated)

Me:

Should I add this threshold calculation fix to the agenda for next week's as a bugfix to consider making for v2024.0.0, separate from the UI for adjusting the size threshold? It will definitely be a breaking change for users, in the sense that processing the same dataset with the segmenter before this change and the segmenter after this change will lead to different outputs - so I think we would need to discuss how to mitigate any breakage for users' existing projects
The alternative is to leave it to be fixed in v2024.1.0 or a later release

Fabien:

yes definitely: but technically it was kind of compensated (althrough very uncorrect in it formulation; it was somehow doing neraly correctly the job); so from the point of view of the user it will not change too much the final results
but when an user asks to ignore 10 micron objects, cutting at 12 micron (or maybe 8 ) this is bad
sorry for this: I thought it was modified since ages....
but this is part of the many “small” imperfections that people ignore to modify so much frequently (and are changing themselfes to big troubles as much as they grow)

[ethanjli]

Here is some discussion on design decisions I had with Fabien:

me:

Here are a few design decisions we will need to make for the UI:

• The name of the "process_min_ESD" setting which we will show to users on the segmentation page
• Should the setting be adjusted as a slider (like ISO), or a text input box (like the "Flowrate" setting), or a number incrementing box (like "Delay to stabilize image" setting)?
• What should happen if the user does not know what value to use for the setting? Should we provide a fixed default value which the user can just accept, or should we leave it blank by default but make it a required setting (so that Node-RED will require the user to enter some number), or should we leave it blank by default as an optional setting and then make the backend automatically decide for the user?

Fabien:

The name of the “process_min_ESD” setting which we will show to users on the segmentation page

= minimal size to segment (equivalent spherical diameter in micron)

Should the setting be adjusted as a slider (like ISO), or a text input box (like the “Flowrate” setting), or a number incrementing box (like “Delay to stabilize image” setting)?

text input box (prefilled to 20 could be an idea)
if we can reload the previous usage filled by the user : even better
(prefilled by 20 at first usage, if the user use a 63 micron mesh: he do not whant to change this value everytimes)