🪸 Coral AI

Goal: reliable, pixel-level underwater segmentation for coral reef monitoring.
We’re focusing on cover estimation, substrate classes, and benthic categories that translate cleanly into survey metrics and reports.

What we’re building

  • Full-frame segmentation for reef scenes (coral types, algae, rubble, sand, etc.).
  • Quality-checked outputs suitable for transect analysis and time-series change.
  • Interoperability with MariMap (GeoJSON masks, tiled overlays, exports).

Assisted labelling in MariMap

We’re adding assisted segmentation to speed up annotation and QA:

  • Smart brush & auto-masks: suggest masks from a click or rough stroke.
  • Edge-aware refinements: quick boundary cleanups with snap-to-edge.
  • Active learning loop: prioritise uncertain regions; improve the model as you label.
  • Batch review: apply/compare masks across transects and image sets.

These tools are designed to support transects & image labelling workflows and to slot into your existing project → site → survey flow.


Try the demo

Direct link: reefsupport-coral-ai-demo.hf.space (opens in a new tab)


Where this is heading

  • Short term: improve class set & colour maps, export masks for MariMap layers.
  • Mid term: assisted labelling in MariMap with smart tools and versioned datasets.
  • Long term: robust reef-wide analytics feeding ESG reporting and biodiversity restoration planning (incl. potential credit methodologies).

Stay connected