OrganicVision
DocumentationApplication TourGet Involved

OrganicVision

OrganicVision is available under [Apache License 2.0](LICENSE). Public download and reuse follow that license; prospective contributors should read CONTRIBUTING.md and provide the background, interest, and participation information requested on the Get Involved page.

Phase 12 physical validation is currently Blocked / Not tested, not Passed. The versioned [PhysicalValidation](PhysicalValidation/validation-manifest.json) package contains complete result records with zero physical evidence. See Testing OrganicVision With Your Hardware and Open-Source Readiness.

Phase 11 adds integration hardening and an explicitly enabled physical-device diagnostic mode. Automated checks use deterministic fixtures and never claim physical camera, sensor, radio, GoPro, DJI, thermal, battery, accessibility, or signed-installation validation. Start with Getting OrganicVision Onto Your Devices, Physical Device Diagnostics, and the Release Checklist.

OrganicVision is a native Swift multi-device computer-vision research platform. Phases 1–9 provide capture, secure connectivity, datasets, immutable analysis, review/annotation, training export, and reproducible perception experiments. Phase 10 adds immutable external-media import, explicit GoPro HERO10/DJI Mini 4 Pro/generic support levels, chapter and telemetry parsing, offset/drift synchronization, multi-camera sessions, camera profiles/calibration, derived proxies, and recoverable validation. DJI flight control is not implemented or permitted.

Simulated iPad Research Dashboard
Figure 1. iPad — Phase 6 Research Dashboard using deterministic simulated research data. Simulated research data.

See Assisted Annotation, How AI Assistance Helps Human Review, and the updated Application Tour.

OrganicVision is a research recorder. Mount and configure it safely before travel. Never hold or operate the application manually while driving.
The Research Console is for a passenger or parked use only. Its live connection is not safety-critical and must never affect safe vehicle operation.
Object labels, boxes, tracks, and environment scores are experimental and can be wrong. OrganicVision is not autonomous driving, collision warning, navigation, certified perception, or a safety device. Human review is required.

Open the workspace

  1. Clone or copy the repository to the M1 Mac.
  2. From the repository root run xcodegen generate if OrganicVision.xcodeproj is absent or project.yml changed.
  3. Double-click OrganicVision.xcworkspace, or run open OrganicVision.xcworkspace.
  4. Do not open Package.swift or the generated project as a substitute for the workspace.

Configure signing

  1. In Xcode, select the blue OrganicVision project in the navigator.
  2. Select each app target: OrganicVision iPhone, OrganicVision iPad, and OrganicVision macOS.
  3. Open Signing & Capabilities, enable Automatically manage signing, and select your Apple Development team.
  4. If a bundle identifier is already registered to another developer, change org.organicvision.iphone, org.organicvision.ipad, and org.organicvision.macos to unique reverse-DNS identifiers. To make the changes durable, edit PRODUCT_BUNDLE_IDENTIFIER in project.yml, then run xcodegen generate again.

Build each target

Select Product → Scheme and choose the desired app scheme/target, select an appropriate destination, then press ⌘B:

The shared OrganicVision scheme builds all three targets together. Command-line validation commands are in Documentation/BuildAndSigning.md.

Run in a simulator

Choose an iPhone or iPad simulator from Xcode’s destination menu, select the matching app target, and press ⌘R. The iPhone simulator shows a conspicuous SIMULATED CAMERA preview and creates a deterministic playable video plus simulated sensor records. Its manifests are marked simulatedCamera and cannot be mistaken for physical recordings.

Install on a physical iPhone or iPad

  1. Connect the unlocked device to the Mac by USB, accept Trust This Computer, and enable Developer Mode under Settings → Privacy & Security → Developer Mode if prompted.
  2. Choose the device in Xcode’s destination menu.
  3. Select OrganicVision iPhone for an iPhone or OrganicVision iPad for an iPad.
  4. Confirm the correct development team and a unique bundle identifier, then press ⌘R.
  5. If iOS asks you to trust the developer certificate, open Settings → General → VPN & Device Management, select the developer identity, and tap Trust. Re-run from Xcode.

Common signing fixes: select a valid team, use unique bundle identifiers, sign in under Xcode → Settings → Accounts, reconnect and trust the device, enable Developer Mode, and use Product → Clean Build Folder after correcting signing. See Documentation/Troubleshooting.md for specific error patterns.

On first physical recording, grant camera, location, and motion access when requested. Enter a session name, optionally add notes, choose quality, and tap Start Recording. Tap Stop Recording before leaving Capture. The Sessions tab plays video, displays metadata, prepares a shareable .organicvision raw package, and confirms before deletion. Packages remain offline under the app’s Application Support directory.

For Phase 3, place both devices on the same trusted local Wi-Fi network. Open Secure iPad Connection on iPhone and OrganicVision iPad on iPad. Select the Camera Node, compare the six-digit code on both screens, and approve only if it matches. Remote recording requires a separate first-use approval on iPhone and can be disabled. Forgetting a device revokes trust. See Using the Research Console.

Tests

Run swift test. The suite covers all earlier phases plus pinned experiment configuration, approved ground truth, deterministic splitting and extraction, transparent fusion, abstention, evaluation metrics, feature-cache integrity, pause/resume/recovery, classical-ML export, and reproducibility reports.

Experimental perception research

Launch iPad or macOS with -organicvision-research-mode to open the Phase 9 experiment workspace. The deterministic fixture and all demonstration metrics are explicitly labeled simulated; they verify software behavior and reproducibility, not real-world accuracy, statistical significance, causation, or autonomous-driving suitability. Start with Research Experiments, What OrganicVision Is Testing, and Statistical Limitations.

External cameras and video

Launch iPad or macOS with -organicvision-external-import-mode for the Phase 10 import workspace. Explicitly select a GoPro SD-card/copy, DJI video plus optional exported sidecars, iPhone recording, or generic MP4/MOV. Inspection, hashes, preservation strategy, missing telemetry, synchronization, calibration, and warnings are reviewed before finalization. Deterministic fixtures do not validate physical GoPro/DJI behavior. See Using Your Existing Cameras, External Camera Import, GoPro HERO10, and DJI Mini 4 Pro.

Assisted annotation

Launch an annotation screen with -organicvision-annotation-mode. Suggestions always remain separate from verified annotations until a reviewer explicitly accepts or edits them. Phase 8 uses a deterministic simulator fixture; it does not bundle or download a trained assistance model. Model and plugin output must include versioned provenance and pass bounded validation. See Annotation Suggestions, Advanced Segmentation, and Adding Assistance Models.

Dataset catalog

Open the Catalog tab on iPhone or iPad; the macOS app opens directly into its desktop catalog. A session package contains captured files and its immutable manifest. Its catalog record points to that package and stores editable names, notes, tags, favorites, review state, collections, provenance, and integrity status. Imports and transfers remain untrusted until validation succeeds. The visible Simulate Import action on iPad/macOS creates only a clearly labeled simulated catalog record.

Phase 4 core services accept validated OrganicVision session directories and produce full or metadata-only versioned dataset bundles. Native Files/share-sheet/ZIP/drag-and-drop/raw-video adapters are defined as future platform work and are not claimed complete. See Dataset Catalog, Session Import, and Session Transfer.

Vision analysis

Open Analysis on iPhone, iPad, or macOS to run the deterministic simulator detector and inspect boxes, track IDs, confidence, and experimental scores. The fixture is project-owned and needs no downloaded model. A real compatible Core ML detector must be imported deliberately with its SHA-256 hash and license metadata; OrganicVision never downloads models automatically.

Live analysis can be disabled while recording continues. Offline work creates a new immutable run under the session’s analysis directory, writes observations incrementally, supports checkpoint/pause/resume/cancel/recovery, and never modifies the original video. Start with How OrganicVision Sees Video, Vision Analysis, and Phase 5 Limitations.

Searchable HTML documentation

The self-contained [HTML Documentation](HTML%20Documentation/) folder can be uploaded directly to shared LAMP hosting. It needs no PHP or database. Search runs locally in the browser from search-index.json. After changing README or any file in Documentation, regenerate the site from the repository root:


python3 Tools/generate_html_documentation.py

Get involved

Developers, researchers, students, testers, technical writers, and computer-vision specialists are invited to participate. Read Get Involved to learn where help is valuable and how to request code access.

Research dashboard and application tour

Phase 6 adds the adaptive iPad and macOS Research Dashboard, synchronized Session Review, object and track inspection, neutral analysis-run comparison, bookmarks, warnings, accessible Swift Charts, and deterministic documentation screenshots. See the non-technical Research Dashboard Overview and OrganicVision Application Tour.

Start with Getting Started, Using the Camera Node, iPhone Camera Capture, and Dataset Catalog.