AI species identification
Upload an underwater photo. Get a species ID with a confidence score. Formal identification takes 13.5 years per species today. Our models train on FathomNet's 100K+ labeled images. They hit 95-98% accuracy on known species. Unrecognized finds get flagged for human review.
Download and explore FathomNet's 100K+ labeled images and VIAME datasets. Assess taxonomy coverage, image quality, and gaps. This is the foundation everything else trains on.
Integrate a vision-language model pipeline using OpenAI and Gemini models for deep analysis of unknown or unclassified species. When the classifier flags something it can't identify, the VLM generates detailed morphological descriptions and suggests candidate taxa for taxonomist verification.
v1 classification model trained and validated on FathomNet imagery. Hits 95%+ accuracy on known species. Anomaly detection flags unknowns for human review and also self resolves with VLM integration. Future iterations will expand taxonomy coverage and improve edge-case handling.
Build the upload-and-identify interface at app.theunfathomed.org. Drag in a photo, get a species ID with confidence score and location context.
Open the platform to researchers, divers, and citizen scientists. Real-world testing across diverse ocean environments proves accuracy.
We need labeled underwater imagery, code contributions on GitHub, and marine biologists for taxonomist review.