AI Models

The largest, highest-accuracy variant of MegaDetector v6. Locates animals, people and vehicles in camera-trap photos. Use this for offline review or batch projects where every detection counts.

Compact variant of MegaDetector v6 — runs faster than the extra-large model with a small accuracy cost. Good default for real-time projects on modest hardware.

Detector trained on European camera-trap data. Locates animals in the frame; pair with a Deepfaune classifier to identify species.

The previous-generation MegaDetector. Kept for projects that have validated against v5 outputs and don't yet want to switch to v6.

Aerial-survey detector for large African mammals (elephant, buffalo, kob, topi, warthog, waterbuck, plus background). Built for densely-populated savanna scenes captured from aircraft.

Single-class aerial detector that flags any animal in the frame. Use for broad presence/absence aerial surveys where species ID isn't required.

Regional YOLO model for sub-Saharan camera-trap wildlife. Returns a bounding box + species label in one pass — no separate classifier needed.

Regional YOLO model for European camera-trap wildlife. Returns a bounding box + species label in one pass — no separate classifier needed.

Regional YOLO model for North American camera-trap wildlife. Returns a bounding box + species label in one pass — no separate classifier needed.

Species classifier tuned to Amazon rainforest fauna. Higher precision than the global classifier on regional species.

Species classifier trained on the Snapshot Serengeti dataset. Strong on East African savanna species.

Regional classifier from the Deepfaune family, tuned to New England (US) wildlife. Pair with the Deepfaune detector.

European species classifier from the Deepfaune project. The standard pairing for the Deepfaune detector across European camera-trap surveys.

Very-large global classifier — covers ~2,500 species. Use when the project covers diverse regions or when no regional classifier exists for your area.

Bird-focused acoustic detector. Slides a 1-second window over the audio file and emits per-window class probabilities. Use for dawn-chorus surveys, bird-call presence/absence, or any project where you want a smaller, faster audio model than Perch.

Very-large global acoustic classifier covering ~15,000 bird species + some non-bird wildlife. Best accuracy on diverse international datasets but slower than AudioBirds; pick this when species coverage matters more than speed.

Stage 1 of the DCLDE 2026 killer-whale cascade — a binary orca-vs-rest screener for continuous hydrophone audio. Designed for Pacific Northwest monitoring corpora (Salish Sea, North BC); pair with the Orca Ecotype Classifier (Stage 2) to identify ecotype on orca-positive windows.

Stage 2 of the DCLDE 2026 killer-whale cascade — a temperature-calibrated 5-class ecotype classifier covering Southern Resident (SRKW), Transient/Bigg's (TKW), Southern Alaska Resident (SAR), Northern Resident (NRKW), and Offshore (OKW) killer whales. Best used on orca-positive windows surfaced by the Orca Detector.