OrganicVision Application Tour

Human-recorded physical results

Diagnostic screens guide tests, while PhysicalValidation/ records the human outcome. Symbols and text are paired: ✓ Passed, ⚠︎ Passed with warnings, ✕ Failed, ⊘ Blocked, ↷ Skipped, — Unsupported, and ○ Not tested. Current physical records contain only Blocked and Not tested states.

Physical-device readiness

The diagnostic route presents guided camera, sensor, storage, power, networking, transfer, external-media, playback, and accessibility checks. Every demo view displays Simulated Diagnostic Data. iPad and macOS also provide release summaries and redacted diagnostic review.

Phase 10 cameras and video sources

The external-import workspace guides authorized selection, inspection, GoPro chapters/GPMF, DJI sidecars, telemetry mapping, manual synchronization, calibration profiles, multi-camera playback, proxy generation, and validation/audit. iPhone shows compact transfer and warning status. All images below use deterministic fixtures.

iPad External Import Wizard — Figure 1. iPad — Phase 10 authorized external-media selection. Simulated research data.

iPad Manual Synchronization — Figure 2. iPad — Phase 10 offset and drift anchors. Simulated research data.

macOS Import Audit Report — Figure 3. macOS — Phase 10 source and telemetry audit. Simulated research data.

Phase 9 research experiments

The iPad and macOS Research workspace provides experiment browsing and building, confusion matrices, feature ablation, error analysis, neutral comparison, summaries, reproducibility, feature tables, threshold and calibration views, generated reports, and cache management. iPhone supplies a lightweight monitor. All examples are deterministic simulated data, not evidence of field performance.

iPad Experiment Builder — Figure 4. iPad — Phase 9 pinned experiment configuration. Simulated research data.

iPad Feature Ablation — Figure 5. iPad — Phase 9 fixture feature-ablation results. Simulated research data.

macOS Research Report — Figure 6. macOS — Phase 9 generated research report. Simulated research data.

All images in this tour use Simulated Research Data. They do not represent a real driving study, a real location, or production model accuracy. OrganicVision is a research platform—not autonomous driving, collision warning, or a safety system. Review while parked, by a passenger, or from prerecorded sessions.

iPhone Camera Node

The iPhone remains a focused recorder. It combines camera and sensor status with lightweight live analysis and encrypted Research Console status.

Simulated iPhone live object detections — Figure 8. iPhone — Live detections and persistent track identifiers over simulated imagery. Simulated research data.

iPad Research Console

The iPad is the portable research dashboard. The overview summarizes sessions, storage, analysis runs, objects, tracks, review work, charts, and recent warnings without loading private data.

Simulated iPad Research Dashboard overview — Figure 9. iPad — Research Dashboard overview with category and environment-composition charts. Simulated research data.

Simulated iPad Live Research Console — Figure 10. iPad — Live Research Console with preview, object list, connection, storage, battery, and sensor status. Simulated research data.

macOS Research Workstation

The Mac provides a larger review surface for catalogs, charts, object searches, track details, comparison, and export preparation.

Simulated macOS Research Dashboard — Figure 11. macOS — Research Dashboard rendered at the standard documentation size. Simulated research data.

Simulated macOS object index — Figure 12. macOS — Searchable object index and detailed result inspection. Simulated research data.

Recording a session

The source video remains immutable. Recording, sensor collection, preview transport, and derived analysis results keep their existing Phase 2–5 boundaries.

Simulated iPhone session browser — Figure 13. iPhone — Session browser containing only deterministic simulated sessions. Simulated research data.

Simulated iPhone session summary — Figure 14. iPhone — Session summary with recording, analysis, and sensor availability. Simulated research data.

Reviewing objects

The session-review screen synchronizes overlays, frame position, presentation time, session-relative time, timeline events, and sensor availability. Frame stepping is available while paused at 0.25×, 0.5×, 1×, 2×, and 4× playback settings.

Simulated iPad session review — Figure 15. iPad — Session review with bounding boxes, playback controls, and an interactive timeline. Simulated research data.

Simulated iPad object inspector — Figure 16. iPad — Object inspector with labels, confidence, environment scores, and human review state. Simulated research data.

Simulated macOS track detail — Figure 17. macOS — Persistent track detail with confidence trend and centroid path. Simulated research data.

Comparing analysis runs

Comparisons use neutral language. A difference, agreement, disagreement, added detection, missing detection, or changed category does not automatically identify the correct result.

Simulated iPad analysis-run comparison — Figure 18. iPad — Neutral comparison of two deterministic runs from the same source session. Simulated research data.

Simulated macOS analysis-run detail — Figure 19. macOS — Analysis-run model, configuration, frame, result, performance, and warning details. Simulated research data.

Understanding warnings

Warnings combine capture, storage, sensor, network, analysis, model, integrity, thermal, import, and export events. Experimental classification uncertainty remains a review concern rather than a system failure.

Simulated iPad warning center — Figure 20. iPad — Warning center sorted by severity with suggested actions. Simulated research data.

Managing datasets

Collections organize sessions without changing immutable packages. Storage summaries separate source video from sensor, analysis, thumbnail, and export assets.

Simulated iPad dataset collection — Figure 21. iPad — Dataset collections and session-storage summaries. Simulated research data.

Simulated macOS storage management — Figure 22. macOS — Storage-management summary preserving integrity and active-session protections. Simulated research data.

Correcting and Preparing Training Data

Human work lives in a versioned annotation layer, while source video and AI results remain unchanged and comparable.

Human annotation workspace — Figure 23. iPad — Touch-first box and polygon annotation workspace with explicit human layer. Simulated research data.

Review queues, track correction, autosave, revisions, and reviewer comments prepare approved examples without claiming that review guarantees correctness.

Figure 24. iPad — Training dataset builder with session-aware and track-aware splits. Simulated research data.

Exports are validated before finalization and exclude sensitive metadata by default. Visual privacy and licensing review remain the exporter’s responsibility.

Figure 25. macOS — Export validation report and privacy warning. Simulated research data.

Reviewing Assistance and Segmentation

Suggested geometry uses its own labeled layer. A reviewer must accept, edit, reject, or defer it; no suggestion silently replaces a human record.

Assisted annotation workspace — Figure 26. iPad — Separate human and simulated suggestion layers with explicit review controls. Simulated research data.

Mask tools expose brush size, opacity, point prompts, and text states without requiring color perception or Apple Pencil.

Point-prompt mask refinement — Figure 27. iPad — Positive and negative prompts over an unreviewed simulated mask. Simulated research data.

Queues explain why work was prioritized, while quality reports show which findings block export. Multiple model outputs are compared neutrally.

Annotation quality report — Figure 28. macOS — Blocking and advisory quality findings from simulated data. Simulated research data.

Model disagreement — Figure 29. macOS — Neutral model-output comparison without an automatic winner. Simulated research data.