L.I.O.N. model observability

Model Graphs And Detector Summary

This page centralizes the core training and validation plots used to tune the hosted invasive-species lane and reef-health detector strategy.

Back To Dashboard

Detector Lanes

Hosted lane

Lionfish Watch

Hosted Roboflow invasive-species detector for lionfish images and video.

Short labelLionfish
Species classes listed1

Hosted lane

Crown of Thorns

Hosted Roboflow detector for crown-of-thorns starfish so coral-predator outbreaks can be spotted early.

Short labelCOTS
Species classes listed1

Remote or local lane

Reef Health Suite

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.

Short labelReef Suite
Species classes listed13

Training And Validation Graphs

Precision-Recall Curve

Precision-Recall Curve

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

Precision vs Confidence

Precision vs Confidence

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

Recall vs Confidence

Recall vs Confidence

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

F1 vs Confidence

F1 vs Confidence

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

Confusion Matrix

Confusion Matrix

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

Training Loss

Training Loss

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