One deep-sea dive line, end to end: a Google-Maps-style seafloor mosaic with automated polymetallic-nodule detection and real-world sizing, a megafauna expert-review queue, and the analytics that fall out of it — every number computed from the sample imagery, de-duplicated across the 70% frame overlap and sized from the georeferencing. Explore the map, then scroll for the dashboard.
Every number below is computed from the sample imagery by the pipeline — de-duplicated across the 70% frame overlap, sized in real-world units from the georeferencing.
Equivalent circular diameter (cm), unique nodules only.
Raw detections vs unique nodules after geometric dedup — the "70% overlap" answer, quantified.
Survey line split into 6 segments — nodule density (per m²) along track.
Raw (all frames) vs owned (counted once). Consistency = stable capture.
Low-confidence candidates are never auto-labelled — they route to a marine biologist. This nodule-pavement line yields few candidates, all in review.
| # | Candidate taxon | Conf. | Status |
|---|
Frames with density >1.8σ from mean, plus every fauna candidate — the "flag for expert review" surface.
| Item | Detail | Value |
|---|
Machine-readable outputs — drop straight into QGIS / ArcGIS / pandas.
Sample exports from this pilot line. The full pipeline also emits a UTM GeoJSON / GeoPackage of every sized nodule for direct GIS ingest.