Get Involved

OrganicVision is an open research project exploring how cameras, local artificial intelligence, sensor data, and computer-vision techniques can work together to better understand recorded environments.

The project is being developed as a practical research platform, not as a finished autonomous-driving product. It collects video, analyzes visible objects, organizes results, and makes those results easier to review and compare.

If you are interested in reviewing the code, contributing to development, testing the applications, improving the research methods, or helping evaluate results, email Herb Trevathan to request code access. Please include a short explanation of your background, your area of interest, and how you would like to participate. Access and setup instructions can then be provided.

What OrganicVision Does

OrganicVision uses an iPhone as a mobile camera and sensor source, an iPad as a larger research console, and a Mac for dataset and offline-analysis work.

The iPhone can:

• Capture live video.
• Record location and motion information.
• Save synchronized research sessions.
• Participate in secure local preview and remote-console workflows.
• Run deterministic local computer-vision demonstrations.
• Display bounding boxes, labels, confidence values, and tracking identifiers.

The iPad can:

• Display the research-console preview foundation.
• Display shared analysis results on a larger screen.
• Show detected and tracked objects.
• Display recording, sensor, storage, battery, thermal, and connection information.
• Review and organize research sessions.
• Receive finalized session packages through the resumable transfer foundation.

The macOS application can:

• Import and organize supported session directories.
• Run deterministic offline-analysis workflows over injected frames.
• Review analysis results and compare multiple runs.
• Search detected objects and tracks.
• Inspect experimental natural-versus-manufactured scores.
• Export reports, logs, metadata, JSON/JSON Lines results, and derived export fixtures.

Physical-device networking, production video-frame decoding, real-model inference, and fully rendered annotated-video encoding still require further implementation or validation. The documentation identifies these boundaries explicitly.

Supported Video Sources

OrganicVision currently supports iPhone capture, recorded OrganicVision session packages, imported-session foundations, and deterministic simulated footage. The architecture is designed to add raw imported videos, GoPro footage, DJI drone footage, and other external recordings through explicit adapters.

External recordings do not need GPS or motion data. OrganicVision reports which information is available and must never fabricate missing telemetry.

What the Analysis Produces

The implemented shared output models and deterministic workflows can produce:

• Object labels and alternate label candidates.
• Normalized and pixel-space bounding boxes.
• Original model-confidence values.
• Persistent tracking identifiers and track histories.
• Object and track timelines.
• Category and session/run summaries.
• Searchable object indexes.
• Analysis performance, compute configuration, warnings, and pipeline/model versions.
• Incremental JSON Lines detection and track records.
• Experimental naturalness, manufacturedness, and unknown scores.
• Plain-language score evidence and explanations.
• Baseline geometry, texture, frequency-pattern, and shape-complexity measurements.
• Motion-evidence extension fields.
• Deterministic simulated segmentation-mask records.
• Lightweight manual verification states and notes.
• Cancelable derived-video export fixtures with overlay sidecar metadata.

These are experimental research outputs. Production Core ML inference, calibrated confidence, real segmentation, complete annotated-video rendering, and scientific validation remain future work.

Areas Where Help Is Valuable

iOS and macOS Development

• Swift and SwiftUI.
• AVFoundation, Vision, Core ML, VideoToolbox, and Network.framework.
• CryptoKit, Core Location, and Core Motion.
• Performance profiling, accessibility, and interface testing.

Computer Vision

• Object detection and tracking.
• Image segmentation and optical flow.
• Feature, frequency, texture, and geometry analysis.
• Confidence calibration.
• Natural-versus-manufactured classification and dataset evaluation.

Research and Validation

• Designing controlled, reproducible experiments.
• Measuring false positives and false negatives.
• Testing different environments, lighting, weather, seasons, and camera sources.
• Improving score explanations and documenting limitations.

Dataset and Annotation Work

• Reviewing detections, labels, bounding boxes, and tracks.
• Identifying missed or duplicate objects.
• Organizing repeatable validation footage.
• Preparing future reviewed training data.

Documentation and Education

• Improving technical and non-technical explanations.
• Creating diagrams, tutorials, and test procedures.
• Helping new contributors understand the architecture.

Who Can Participate

You do not need to work for a large research laboratory or technology company. Potential contributors include software, electrical, computer-vision, machine-learning, automotive, and robotics engineers; university students and educators; technical writers and testers; photographers, videographers, and drone operators; and people with practical experience evaluating roads, infrastructure, construction, or natural environments.

Contributions should be based on careful testing, documentation, and reproducible results.

How Participation Works

1. Email Herb Trevathan.
2. Briefly describe your background and area of interest.
3. Explain whether you want to review code, test the application, contribute code, provide footage, improve documentation, or evaluate research results.
4. Receive the current project-access and setup instructions.
5. Follow the project’s branch, build, testing, and documentation requirements.
6. Submit proposed changes for review instead of modifying completed phases without documentation.

As the project matures, this may expand into formal contributor guidelines, issue templates, testing protocols, and pull-request procedures.

Current Project Status

OrganicVision is under active development. The current platform includes a modular Xcode workspace; iPhone, iPad, and macOS applications; local session recording; synchronized video/location/motion records; secure local-network foundations; dataset cataloging and integrity verification; deterministic detection; tracking and object indexing; immutable analysis runs; pause/resume/cancel/recovery; experimental evidence fusion; searchable HTML documentation; and automated shared-module tests.

Some capabilities use deterministic fixtures or require physical-device validation. Project documentation should always distinguish what is implemented, simulated, tested, or experimental.

Important Research and Safety Notice

OrganicVision is an experimental computer-vision research platform. It is not a self-driving system, collision-warning system, navigation system, certified automotive product, replacement for driver attention, or safety-critical perception system.

The software must not be operated manually by a driver while a vehicle is moving. Testing should be performed while parked, by a passenger, with prerecorded footage, or in controlled environments. All AI-generated labels, tracking results, scores, and explanations require human review.

Contact

To request access to the code or discuss research participation, email herb.trevathan@icloud.com.

Include your name, relevant experience, area of interest, how you would like to contribute, and—when relevant—the devices or development hardware available to you.

Search topics: get involved, contribute, contributor, research participation, code access, OrganicVision research, testing, computer vision, Swift developer, Core ML, dataset review, documentation, volunteer, collaboration.