Accurate Segmentation of Elongated Hypocotyls
The original model often mislabeled long hypocotyls, cutting them short and throwing off their length. The retrained model segments them correctly.
Root & hypocotyl phenotyping
Plant phenotyping, tracked plant by plant.
A GUI pipeline for segmenting and measuring root and hypocotyl growth from Arabidopsis timelapse plates: image alignment, nnU-Net segmentation, and per-plant growth curves, end to end.
Fork of ngaggion/ChronoRoot2 · maintained at
calvinyong1/ArabidopsisAnalysisV2
plate3 / plant_2 MainRootLength HypocotylLength
The pipeline
Three GUI Apps, Each With Its Own run.py
-
Aligns raw plate timelapses frame to frame before segmentation. Falls back to an identity transform when a reference-frame marker can't be found, so one bad frame doesn't block the whole plate.
-
Runs nnU-Net inference over aligned frames (root, lateral root, hypocotyl, seed, leaf, and petiole classes) with GPU detection and a CPU fallback.
-
Turns segmentation masks into per-plant growth curves (main/lateral root length, lateral count, hypocotyl length) and generates the final report.
Changes from ChronoRoot2
What's New in This Fork
Native macOS App
One installer sets up everything and adds clickable apps to Applications. On Apple Silicon, segmentation runs on the Mac's graphics chip. The original supported Macs only through Docker.
New imageAligner App
Corrects small plate shifts between timelapse photos using printed markers, before anything is measured. A frame without a detectable marker is flagged instead of stopping the run.
Fewer Silent Measurement Errors
Hypocotyls are no longer borrowed from neighboring plants or zeroed before the root is found, and growth charts now work at any photo interval.
// Reference
Documentation
DOC_01 Installation Guide → Step-by-step setup for Windows via WSL, including the WSL1-vs-WSL2 GUI gotcha. DOC_02 Image Aligner → Removes horizontal drift between timelapse frames using an ArUco marker. DOC_03 Segmentation → Runs the nnU-Net model on every frame, then temporally smooths the masks. DOC_04 Single Plant Analysis → Tracks one plant's root over time, measures it, and builds statistical reports. DOC_05 What's New in This Fork → A plain-language rundown of what the plant-recognition model can do now, and how the pipeline got more reliable. DOC_06 Training the Model → Dataset prep, splits, VM training, and checkpoint handling for the segmentation model. DOC_07 Architecture Notes → Seed position vs. root base, class labels, path structure, and how main/lateral classification actually works.