Roadmap

Roadmap

The roadmap prioritizes scientific validation before additional feature breadth.

Priority 1 — Real end-to-end graph validation

• real cross-camera identity association;
• real temporal event linking;
• graph node/edge evaluation;
• controlled real-evidence contradictions;
• camera-removal / missing-source experiments;
• graph-level confidence calibration.

Priority 2 — Remove oracle localization

Replace annotation-derived actor tubes with detector/tracker output and quantify crop/localization failures.

Priority 3 — Stronger comparative baselines

Evaluate the same real evidence under:

• time-only association;
• appearance-only association;
• naive fixed fusion;
• fusion without provenance preservation;
• fusion without uncertainty/missingness handling;
• complete IncidentGraph.

Priority 4 — Generalization

• second independent real multi-camera dataset;
• sensor/modality shift;
• camera clock perturbation;
• missing-camera stress tests;
• cross-site evaluation.

Priority 5 — Scaling

• temporal blocking;
• camera-neighbor constraints;
• approximate-nearest-neighbor candidate retrieval;
• scaling curves versus camera count, tracks and event proposals.

Priority 6 — Learned fusion while preserving auditability

Explore learned heterogeneous-graph or cross-modal fusion modules without weakening the provenance, contradiction and missing-evidence contracts.