ReefSpotterby VICAR

Welcome, Friend

Please select an option:

4,287 reef observations submitted by the community to date

About this tool.

About this tool

This tool lets managers and community members photograph a reef organism and get a fast, AI-assisted read on what it is and how it is doing. Every submission helps get a current state of reef condition, flags reefs worth a closer look, and notifies us about things to be on alert about.

How it works. Choose what you saw, upload a photo and outline your subject, and a matching AI model returns an identification in seconds. Your photo is processed in your browser.

  • Detect coral bleaching and disease early, at the colony scale
  • Map reefs and their inhabitants
  • Track emerging invasive species
  • Empower the territory and the community to get involved and make real contributions to saving our coral reefs!

Support. Built by the VICAR Lab and supported by grants from [Grant A] and [Grant B] (2027 to 2029).

How we use your entries

  • First response. You get an instant, AI-assisted read on what you photographed. Learn more
  • Aggregated with our monitoring. Your entry joins our long-term reef monitoring to build a current picture of reef condition. Learn more
  • Alerts. Entries help us flag bleaching levels and timing (learn more), disease progression (learn more), and invasive spread (learn more).

Your email address is removed within 30 days of a human research expert reviewing the identifications. We only ask for your email in case you submit something truly novel, so we have the chance to ask you for more details. You might find a new species!

All identifications

Every identification submitted, shown on the map and in the table below (made up for this demonstration). Filter by category to see all identifications in any group.

DateCategoryIdentificationLocationSubmitter

About us

The VICAR Lab at the University of the Virgin Islands is part of a large team integrating AI, robotics, and open, reproducible data into coral reef monitoring, so we can see reef change in real time, in finer detail, and across more reefs than we can currently monitor with our team of research divers. Your contributions as citizen scientists are essential to our observations of these reefs. Learn more vicarlab.org

The AI models & metrics

Deep in the weeds: each subject type is screened by its own model on the VICAR Compute Core. The reef fish model reflects our February 2026 report; the other figures are illustrative for models in active development. Each model links to its report, weights, and a way to contribute training imagery.

Reef fish

FrameworkYOLO11-Large · Ultralytics (PyTorch)
Training set~9,900 labeled fish · 34 species & morphotypes
Accuracy90.4% mAP@50 · precision 89.5%, recall 83.5%
Intended useDetect and classify reef fish in survey imagery (screening aid)
LimitationsWeaker on rare morphotypes and look-alike species; USVI imagery only
LicenseCC BY-NC 4.0 (research)
Speed2.67 ms / image (~375 img/sec)
Last updatedFebruary 2026

Coral health & disease

FrameworkYOLO11-seg + SAM3
Training set~14,000 annotated colonies · 4 condition classes
Accuracy87% (illustrative)
Intended useFlag bleaching and disease at the colony level
LimitationsConfuses pale and recently bleached tissue; depth and lighting sensitive
LicenseCC BY-NC 4.0 (research)
Last updatedMarch 2026

Coral species

FrameworkYOLO11 + visual-language (OnDeck)
Training set~12,500 colonies · 7+ species
Accuracy84% top-1 (illustrative)
Intended useIdentify reef-building coral to species
LimitationsSimilar species and partial colonies reduce confidence
LicenseCC BY-NC 4.0 (research)
Last updatedMarch 2026

Sponge

FrameworkYOLO11-Large
Training set~6,200 images · 8 species
Accuracy80% mAP@50 (illustrative)
Intended useIdentify common Caribbean sponges to species
LimitationsGrowth-form overlap among species
LicenseCC BY-NC 4.0 (research)
Last updatedJanuary 2026

Sea fan / octocoral

FrameworkYOLO11-Large
Training set~4,300 images · 7 taxa
Accuracy78% mAP@50 (illustrative)
Intended useIdentify sea fans and other octocorals
LimitationsGenus-level confusion among gorgonians
LicenseCC BY-NC 4.0 (research)
Last updatedJanuary 2026

Macroalgae

FrameworkSAM3 segmentation
Training set~9,000 patches · 5 functional groups
Accuracy82% (illustrative)
Intended useClassify algae into functional groups
LimitationsTurf and crustose forms overlap
LicenseCC BY-NC 4.0 (research)
Last updatedFebruary 2026

Invasive species

FrameworkYOLO11 binary detector
Training set~5,000 images · Xenia / Unomia
Accuracy93% recall (illustrative)
Intended useEarly detection of pulsing soft corals (Xenia / Unomia)
LimitationsLook-alike native soft corals
LicenseCC BY-NC 4.0 (research)
Last updatedApril 2026

We appreciate your contribution and curiosity.

Please tell us what you think you saw. We will run the matching model.

What did you see?

Pick the closest match and we will run that model.

Upload a photo & outline your target.

Box your subject · required
Draw a box around your subject (required) Even with just one subject in the frame, outline it. This tells our AI model exactly what to read and makes it work better. Click and drag on the photo.
For a good photo Subject filling the frameSharp focus, steady shotEven, natural light

Where, and who?

Tag your observation. This is required to see results.

Interactive map needs an internet connection. Type a site name or coordinates below.
We will never email you unless you submit a truly noteworthy image.

Your screening read-out.

Running AI model
Preprocessing image…
0%