Skip to content

Segment benthic imagery

Segmentation creates pixel regions or masks. Use it when the geometry of a colony, organism, or substrate patch matters.

A completed point-supervised segmentation job with image and mask counts

Choose a starting mode

Mode Best for
Empty start Manual tracing or an unconstrained mask workflow
Point-seeded masks Turning reviewed points into candidate regions
Point-guided suggestions Interactive refinement around provided points
CSV points A prepared external point file with known coordinates

When a CoralNet-attached visit offers Use imported CoralNet points, the CSV input is not required. Imported points seed a workflow; they do not become mask truth automatically.

Review masks

  1. Open Data and confirm the expected images and point source.
  2. Open Results or the arena.
  3. Inspect boundaries at full resolution.
  4. Correct merged neighbors, holes, missed thin branches, and background leakage.
  5. Confirm the label and mask count before export.

Common ecological pitfalls

  • branching colonies may fragment into multiple masks
  • touching colonies may merge
  • turf or crust boundaries may be visually ambiguous
  • shadows and color loss can mimic substrate boundaries
  • a point may land on an occluding organism rather than the intended layer

Record a consistent rule for these cases before multiple annotators begin.