Hosted lane
Lionfish Watch
Hosted Roboflow invasive-species detector for lionfish images and video.
L.I.O.N. model observability
This page centralizes the core training and validation plots used to tune the hosted invasive-species lane and reef-health detector strategy.
Back To DashboardHosted lane
Hosted Roboflow invasive-species detector for lionfish images and video.
Hosted lane
Hosted Roboflow detector for crown-of-thorns starfish so coral-predator outbreaks can be spotted early.
Remote or local lane
Marine-detect-style paired suite for fish, invertebrates, mega fauna, and rare species. It uses a remote Python service when MARINE_DETECT_API_URL is configured, or falls back to the local YOLO runner.

Shows the precision and recall tradeoff. Curves closer to the top-right indicate better class separation.

Helps select confidence thresholds that improve precision while controlling false positives.

Highlights how sensitivity changes as confidence increases, useful for mission-specific threshold tuning.

Shows the confidence band where precision and recall are most balanced.

Summarizes class-level mistakes so weak labels and overlapping classes can be corrected.

Tracks optimization stability during training and indicates whether convergence is still improving.