Field Geometry
From segmented image fields to a geometric distribution—with calibrated Feret measurements, statistical moments and confidence intervals that respect independent specimens.
Calibrated object geometry
Polygon boundaries or labeled images. Independent X/Y calibration. Exhaustive polygon Feret extrema and analytic mean width, with convex-hull area and perimeter.
Distribution and uncertainty
Raw and central moments through configurable order, dispersion, skewness, kurtosis and quantiles. Default 95% pointwise percentile intervals resample whole specimens.
Results you can inspect
CSV tables, a readable HTML report, frozen inputs, explicit status flags and SHA-256 receipts. A Make grammar exposes the same Python operators used by the CLI.
Validate. Measure. Estimate. Verify.
make doctor
make validate DATA=my.dataset.json
make measure DATA=my.dataset.json
make run DATA=my.dataset.json OUT=runs/case01
make verify OUT=runs/case01
make replay OUT=runs/case01Python 3.11+ and GNU Make. Installation, input contracts and labeled-image examples are included in the README. The ZIP includes a completed synthetic demonstration.
A tested reference implementation.
The bundled synthetic demonstration contains 12 specimens, 36 fields and 180 objects. It produces 140 statistics with 2,000 specimen-cluster bootstrap replicates. A fresh replay reproduces the scientific output files byte for byte in the tested environment.
Tests cover analytic shapes, anisotropic calibration, rotation and translation, independent directional-reference parity, known moments, variance decomposition, specimen weighting, duplicate-object pseudoreplication, image-edge refusal and receipt tamper detection.
The claim travels with the result.
- Observed 2D sections or projections. No automatic reconstruction of 3D particle populations.
- Pointwise, approximate bootstrap intervals. Joint simultaneous coverage and empirical coverage calibration are not established.
- Specimens are the independent resampling units. One specimen yields descriptive estimates with unavailable population CIs.
- Angular distribution moments use finite quadrature. Sampling CIs do not include segmentation, calibration or angular discretization error.
- Computational validation uses synthetic fixtures. Biological validation and parity with the separate QuPath production kernel are not claimed.
This release establishes the reference measurement and inference contract. QuPath/Fiji adapters, model-specific 3D inversion, simultaneous intervals and adaptive field acquisition are future extensions.