Shaded value = in the CV mean (shade deepens with cover) Grey italic = shown for comparison but left out of the CV mean: the gold standard and manual CoralNet annotations (human reference values, not models), Coralscapes (not trusted), and any value that reuses the reference's own data n = frames behind the value — = not run or not applicable (see each region's notes) Site names and method names link out (new tab)

Coral Cover: Saba 2026

4 sites · 342 photos

No gold standard.

SiteDateClaudeDINOv3Coralscapes v1Coralscapes v2CoralSCOPMean ± SD of CVRange
Ladder Labyrinth2026-02-218.9n=8218.3n=8212.7n=8215.4n=8222.3n=7816.5 ± 6.98.9–22.3
Man-of-War Shoals2026-02-213.7n=10219.8n=10216.0n=10125.1n=10129.3n=9817.6 ± 12.93.7–29.3
Tent Drift2026-02-214.3n=7317.7n=7311.7n=7314.5n=7322.1n=6314.7 ± 9.34.3–22.1
Tent Shallow2026-02-223.0n=8515.0n=857.9n=8511.7n=8521.3n=7213.1 ± 9.33.0–21.3
All sites (unweighted mean)5.017.712.116.723.815.5 ± 9.65.0–23.8
Notes for Saba 2026 (11)
  • [a] PENDING Anderson's annotations (not hand-annotated yet; 'I'll do them later', 2026-09-29).
  • [b] PENDING Anderson's annotations (not hand-annotated yet; 'I'll do them later', 2026-09-29).
  • [c] Value read from cv_method_tables.json (injector output, generated 2026-09-28 22:33); the scope text there gives denominator and frame set.
  • [d] DINOv3 (in-house, frozen Alor/Indonesia model) = mean hard-coral probability over 20 synthesized points per photo, NO water/fish class, so not a benthic denominator (from the map build's hard_coral_cover_by_site.csv).
  • [e] Coralscapes (v1 EPFL-ECEO/coralscapes-vit-b-dpt; v2 coralscapesv2-dinov3-vitb-lora-dpt) is not trusted (Anderson): shown, never averaged. Area fraction, not a point count.
  • [f] CoralSCOP = per-frame sum of colony-mask area fractions (zero-filled; overlapping masks double-count), averaged over frames with an ok row. Area index, not a point count: denominator differs from any point-count reference.
  • [g] Not applicable: no edited set exists (backlog 09-25: 0 edited frames).
  • [h] Claude = blind 7-call point count on the UNEDITED photos, 20 points per photo, (HC+BL)/(20 − WATER − FISH − OBST), UNK kept in; mean across photos (arm ledgers, via cv_method_tables.json).
  • Mean/Range: Mean ± SD and Range are over the CV cells that enter the mean (CoralNet where independent, DINOv3, CoralSCOP unedited/edited, Claude). Gold standard, frames manually annotated in CoralNet, Coralscapes and same-measurement cells are shown but never averaged (*).
  • Methods: Claude (Anthropic); DINOv3 (Meta; paper; used here as the frozen backbone of the in-house classifier); Coralscapes v1 and v2 (EPFL-ECEO; dataset paper, code); CoralSCOP (Zheng et al., CVPR 2024); CoralNet.
  • Sites: names drop the region prefix (hover for the full name) and link to the site's CRD page, or to the regional summary page where the site has none. Date = survey date from the map pins (first–last visit); sites are listed earliest first, undated sites last.

All regions

This trip: what has run, photo basis and photo counts

TripDivesPhotosReference / transectsManual pointsCoralNetClaudeDINOv3Coralscapes v2Coralscapes v1CoralSCOPTable
Saba 20264379—pending Anderson's annotationspending Anderson's annotations342 UNEDITED - full-set arm (saba_statia_sxm, complete)342 UNEDITED341 UNEDITED341 UNEDITED325 ok UNEDITEDincluded