Changelog

Dated log of fixes applied to the analysis pipeline (singlePlantAnalysis/analysis/). This page mirrors CHANGES.md. Keep them in sync, or replace this content with an include of the real file.

2026-09-03: analysis/imageUtils/seg.py

Hypocotyl component selection no longer borrows a neighbor’s blob. Added a hard horizontal-distance gate (max_horizontal_distance=80) to hypocotyl anchor selection. Previously, if a plant had no hypocotyl blob yet, the nearest available component (even a distant neighbor’s) was accepted regardless of distance. Horizontal-only distance was chosen over Euclidean because plants are plated in a row and a hypocotyl grows vertically: 80 is half the ~170px minimum seed spacing, minus a margin for blob width.

2026-08-24: Environment Pinning, 3_generateReport.py

Report generation no longer crashes on fresh installs. An unpinned transitive dependency (multimethod) shipped a change that broke scikit-fda’s operator registration with a metaclass-conflict error at import time, taking down report generation entirely, even for configs with FPCA disabled. Fixed by pinning multimethod<2.0.2 and making the FPCA import lazy and failure-tolerant.

2026-08-19: analysis/dataWork.py

Growth charts respect the real capture interval. The hourly reconciliation formula assumed a fixed 15-minute interval; generalized to (N_exp - 1) * timeStep // 60 + 1.

Fixed a medfilt zero-padding artifact. scipy.signal.medfilt implicitly zero-pads past the array boundary, flattening the last few points of every growth curve. Replaced with a centered rolling median that shrinks its window near the edges instead of padding with fabricated data.

2026-08-19: analysis/plantAnalysis.py, analysis/graphUtils/save.py

HypocotylLength no longer zeroed before root detection. Hypocotyl length is computed independently of root graph success, but every early-return path in the “find first valid root structure” loop hardcoded saveProps(..., 0, 0), discarding the already-computed value. The real value now survives all four early-return paths.

2026-08-19: analysis/imageUtils/seg.py

Hypocotyl selection anchored on seed position instead of largest blob. Component selection previously picked whichever class-4 blob had the largest pixel area, with no awareness of which plant it belonged to: a diagonally grown root could widen the ROI enough to catch a neighbor’s larger hypocotyl. Selection now scores candidates by proximity to fixed_seed_position, mirroring the pattern already used for root component selection.

2026-06-12: Path Structure Across fileUtilities.py, run.py, 2_postprocess.py, 3_generateReport.py

Dropped a hardcoded multi-Raspberry-Pi path segment that didn’t match the pipeline’s actual single-rig capture setup. Output paths are now Analysis/{experiment}/{cam}/{plant}/... instead of carrying an extra rpi level.

Missing ArUco marker on a plate’s reference frame no longer blocks the whole plate. Alignment failure on the reference frame now falls back to copying raw images through unaligned, so segmentation can still proceed; the segmentation queue UI gained Not Aligned / Partially Aligned states.

2026-05-19: Image Discovery

.tif / .tiff experiments are no longer silently dropped. Image discovery was hardcoded to *.png only. Added loadImageFiles(), which globs .png, .tif, and .tiff together.

2026-04-16: singlePlantAnalysis/run.py

Blank plate/camera/plant/experiment fields no longer corrupt output paths. Fields default to "1" when blank, with editingFinished handlers so a blank field self-corrects as soon as focus leaves it.

---

Still Open

• prune()’s skeleton erosion has no length-awareness: a structure shorter than roughly 2 x num_it can be fully consumed before it regrows.
• The “remove spurious lateral roots at the beginning” logic in dataWork.py isn’t actually scoped to the beginning; it runs across the entire sequence.
• PostProcess_Hour.csv has NaN in 2 of every 3 hourly rows whenever the capture interval doesn’t divide 60 minutes evenly.