ArganEye

Acquire imagery. Inspect tree crowns. Export reviewed maps.

ArganEye is a geospatial research workspace for native-resolution imagery, experimental tree-crown detection, human validation and land-cover exploration. It combines Vue 3, Python, PostGIS, object storage and optional model runtimes.

This sanitized source snapshot excludes private pilot imagery, predictions, tree locations, experiment reports and Git history.

Get started · User guide · Documentation · Integrate a model · Contribute · Roadmap · Changelog

The real Vue interface running against an isolated, read-only synthetic demo. Locations, imagery, heights, counts and decisions in these screenshots are fictional; they are not model-performance evidence or pilot imagery. Reproduce the screenshots.

Start here

What the platform does

• Regions and imagery: draw/import versioned regions; stitch configured XYZ tiles into JPG/PNG; discover and acquire STAC raster assets.
• Experimental crown proposals: CHMv2 canopy heights from overlapping native 512 px RGB crops, followed by height-derived crown boxes.
• Human validation: tree / not a tree / unsure, notes, revision history and estimated tree centers for completed CHMv2 runs.
• GIS products: crown boxes and centers, review CSV, height rasters, imagery downloads and run provenance. Output reference.
• Land cover and use: historical ESA WorldCover 2021 v200 classes and area summaries, plus separate manual land-use annotations.
• Extension points: framework-neutral inference helpers, model metadata registry and worker pipelines. New models require code integration; registration alone does not load a checkpoint or create a UI option.

More screenshots: analysis setup and land-cover summaries

Try the interface

Use Node.js 22 and Python 3.14.8. From the repository root:

npm ci --prefix web
npm run build --prefix web
python3 examples/docs_demo.py

Open http://127.0.0.1:18081/#review. The synthetic demo needs no database, GPU, provider account or model download. It is read only: inference, saving reviews and artifact downloads require the real stack.

For the local platform:

docker compose up --build

Open http://localhost:18080 when /readyz succeeds. Base Compose does not package the optional CHMv2 runtime, weights or host imagery cache. See setup profiles before running crown inference.

Research status

The agreed pilot is the Admine paper study region near Agadir Al Massira Airport, defined by this ROI. The original airport subset and older map-frame approximation have different footprints.

This is a working research platform, with experimental crown proposals. The pilot has not completed a full-region CHMv2 run or an independent crown/species benchmark. A review decision does not establish species or surveyed trunk accuracy. Incomplete reference points cannot determine false-positive rates. Land cover is a historical 10 m product, not classification of the current RGB mosaic.

Current capabilities · Scientific methods · Distribution notes

Develop and contribute

Start with CONTRIBUTING.md. The developer guide explains setup and edit locations; the architecture guide separates implemented boundaries from proposed migrations. Model contributors should run the adapter example and inspect the synthetic output samples before changing job contracts.

Testing · Governance · Community expectations · Security reporting

This copy is prepared for outside contributions without internal datasets or history. Public hosting, repository protections and a verified reporting contact remain owner-controlled setup tasks. Readiness checklist.

Cite this work

See CITATION.cff for machine-readable metadata. GitHub's Cite this repository action provides citation exports once that file is on the default branch (GitHub documentation).

@software{arganeye,
  author  = {Bouhsine, Taha},
  title   = {ArganEye: Geospatial imagery and tree-crown review},
  version = {0.1.0},
  url     = {https://github.com/mlnomadpy/argan-eye-public}
}

Record the exact Git commit used. 0.1.0 is the declared package version; no published release is implied. Cite upstream models, datasets and the region paper separately when used.

License and data

Original project code and documentation are MIT licensed. Model weights, provider imagery, datasets, supplied annotations and other third-party assets retain their own terms. The MIT license does not clear their redistribution or training use. The new documentation screenshots use original synthetic fixtures. See licensing and data boundaries before sharing a fork, container, dataset or export.