Implementation of Geometric Subsegmentation into Feature Detection

[fsenf]

This has been on my list for years now. I am really interested to get your feedback on any aspect.

Content:
Geometric subsegmentation is an iterative method that works with a single threshold and breaks down the field into smaller parts. Its implementation is scale invariant, i.e. feature that appear similar are treated the same way independent of their size (and of course discretization effects destroy this property for very small objects). The method has been applied for analyzing tropical anvil clouds here:

Senf, F., D. Klocke, and M. Brueck, 2018: Size-Resolved Evaluation of Simulated Deep Tropical Convection. Mon. Wea. Rev., 146, 2161–2182, https://doi.org/10.1175/MWR-D-17-0378.1.

Implementation:

• I made a new submodule, here tobac/segmentation/geometric_subsegmentation.py and just copied over my still existing functions. Likely, there is functionality the is duplicate to existing one --> How to deal with this?
• I am a big fan of curvature flow filters. I kept this in here and this adds a dependency to https://simpleitk.org/about.html

Testing:

• unit tests are missing still
• an example notebook shows the principle. This case is indeed not to different to multi-threshold feature detection. Maybe I need to think about a show case that makes the added values more clear ...
• If you like to run it for your one, please care that your env includes simpleitk, e.g. via mamba install simpleitk

This is really an early drafting stage - also about the question how different feature detection pathways might look like. Please be critical and actively discuss.

[fsenf]

@freemansw1 , @w-k-jones , @JuliaKukulies , @kelcyno : ping for you in case you are interested.

[github-actions]

Linting results by Pylint:

Your code has been rated at 8.36/10 (previous run: 8.37/10, -0.01)
_{The linting score is an indicator that reflects how well your code version follows Pylint’s coding standards and quality metrics with respect to the RC_v1.6.x branch.
A decrease usually indicates your new code does not fully meet style guidelines or has potential errors.}

[fsenf]

@freemansw1 , @w-k-jones , @JuliaKukulies , @kelcyno : ping you again. Do you really have no opinion on this development idea?

[w-k-jones]

@freemansw1 , @w-k-jones , @JuliaKukulies , @kelcyno : ping you again. Do you really have no opinion on this development idea?

I've only had time to have a brief look at this, but it appears very nice! I think it is very complementary to Sean's family detection in #551 , it might be a nice addition to add a keyword to allow preserving the original feature labels as families after subsegmentation. Let me know if you would like a more in depth review on this as it is currently, and I can put some time aside for it later this week :)

[codecov]

Codecov Report

❌ Patch coverage is 14.08935% with 250 lines in your changes missing coverage. Please review.
✅ Project coverage is 61.49%. Comparing base (8639ece) to head (0d5138a).

Files with missing lines	Patch %	Lines
tobac/segmentation/geometric_subsegmentation.py	11.24%	229 Missing ⚠️
tobac/feature_detection.py	30.00%	14 Missing ⚠️
tobac/utils/general.py	41.66%	7 Missing ⚠️

Additional details and impacted files

@@              Coverage Diff              @@
##           RC_v1.6.x     #556      +/-   ##
=============================================
- Coverage      64.86%   61.49%   -3.37%     
=============================================
  Files             27       28       +1     
  Lines           4047     4332     +285     
=============================================
+ Hits            2625     2664      +39     
- Misses          1422     1668     +246     

Flag	Coverage Δ
unittests	61.49% <14.08%> (-3.37%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.