Public roadmap

Current foundation — PanorAi 3.8

  • Preserve geometry-v1, the Stable object workflow, typed Pipeline, spherical-operation library, CPU Torch layers, SLIC, morphology and the audited component catalog.

  • Keep Stable and Compatibility 3.x imports working. Experimental pose, stereo, reconstruction, semantics, graph and SLAM surfaces retain their existing evidence gates.

  • Preserve exact artifact identity across source, wheel, sdist, TestPyPI and PyPI. Source-checkout success alone is not release evidence.

Delivered research foundations

The first two ordered capability priorities are now Stable:

  1. Research evaluation: public, geometry-aware metrics, evaluation records and reproducible protocols for spherical algorithms.

  2. Region analysis: topology and measurements that continue the Stable morphology and SLIC foundations into connected regions, distances, ordered contours, medial axes and watershed-style grouping.

Current Experimental development — registration and stitching

The current Experimental capability is:

  1. Registration and stitching: construct one typed panorama from \(N \geq 2\) overlapping images, estimating missing image geometry when it is observable instead of requiring metadata for every image.

The detailed computer-vision research capability roadmap defines scope, API placement, scientific gates and explicit non-goals. PanoramaStitcher now implements the bounded central rotation-only profile with reusable calibration, robust model fitting and public Pipeline Stages. It is not a flagship Application and remains Experimental while global seam/multiband processing, broader camera models, real-image prospective evidence, installed-wheel evidence and independent review remain open.

Beyond those priorities

  • Extend motion and video support with spherical vector fields, optical flow, tracks, motion segmentation and stabilization after the representation and evaluation contracts exist.

  • Mature existing pose, depth, reconstruction and SLAM paths through their prospective scientific gates rather than promoting them by breadth alone.

  • Continue graph association, localization, confidence and query research on spatially independent multi-view corpora.

  • Add photometric, spectral, rendering or learned-model capabilities only when they preserve the package focus and have a clear sphere-specific contract.

Later — PanorAi 4.0

  • Apply the published import-destination map with one coherent migration.

  • Replace implicit global registries with immutable configuration and explicit constructors where evidence supports the change.

  • Consider optional point-cloud I/O adapters and additional graph storage backends only after their portable contracts are established.

Promotion remains evidence-driven. This roadmap does not promote an Experimental surface, authorize a release, or weaken an existing scientific gate.