Changelog

All notable changes to ForceInferencePy are documented here. The format follows Keep a Changelog.

[Unreleased]

Napari plugin (force_inference._napari) exposing the full pipeline as a 6-tab dock widget — Segment, Topology + Curvature, Solve + Stress, Visualise, TimeSeries, and 2.5D / 3D — wiring every pipeline parameter to a Qt control. Segmentation runs in a spawned subprocess to avoid the macOS OMP/PyTorch segfault inside Qt threads. Launch with python -m force_inference._napari.
Batchelor stress crosses rendered as napari Vectors layers (principal axes of the per-cell stress tensor: red = tension, blue = compression).
Portfolio poster generator (scripts/generate_portfolio_poster.py) that renders a recruiter-ready figure from a live pipeline run.

solve_laplace is now called with its real signature in the napari widget (regularization, tension_val, detrend, zero_center, border_margin); curvature is auto-computed when the Young-Laplace solver is selected.

Label-driven topology extraction is 14–21× faster with byte-for-byte identical output (equivalence-tested). Two hot paths each ran a full-image operation inside a per-item loop (O(n_items × H × W)):
- _build_edges_from_corners (topology_label.py) — replaced full-map connected-component labelling per cell-pair with bounding-box-local labelling indexed in a single pass over the corner map.
- _cluster_vertex_corners (topology_label.py) — replaced a full comp_label == cid scan per vertex component with one-pass grouping of component pixels.
- Benchmark harness and baseline/after numbers added under benchmarks/.

Core data structures (force_inference/core.py)

Tissue dataclass — unified container for 2D/2.5D tissue topology (vertices, edges, cell neighbours, label mask).
ForceResult dataclass — stores inferred tensions, pressures, residual, and optional stress tensors.

Segmentation (force_inference/segmentation.py)

segment_grayscale — h-minima watershed pipeline for membrane-labelled images (TIFF, PNG, JPEG).

Topology extraction (force_inference/topology.py, force_inference/topology_label.py)

Skeleton-based topology extraction (extract_topology).
Label-driven topology extraction (extract_topology_label) — handles twin junctions without skeleton merging artefacts; supports stub collapse, tiny-twin promotion, spline resampling, and sub-pixel vertex snapping.

Geometry (force_inference/geometry.py)

compute_curvature — pixel tracing + circle fitting + analytical tangent computation.
map_z_to_vertices — maps brightest-Z position onto tissue vertices for 2.5D stacks.
calculate_batchelor_stress — per-cell 2×2 stress tensor via Batchelor formula.
interpolate_stress_to_grid — Gaussian-weighted coarse-graining of cell stress onto a regular grid.

Solvers (force_inference/solvers.py)

solve_bayesian — Bayesian force inference with automatic μ selection via log-evidence maximisation.
solve_laplace — Laplace-pressure solver with border-cell atmosphere treatment.
BayesianScanResult — structured result for μ-scan output.

Visualization (force_inference/visualization.py)

Examples (examples/)

demo_2d_bayesian.py, demo_laplace.py, demo_stress_analysis.py, demo_25d_stack.py.
Diagnostic scripts: diagnose_junctions.py, diagnose_tif_vs_jpg.py, compare_methods.py.

Documentation