Efficient pipeline for camera trap image review

Beery, Morris, and Yang described the approach behind MegaDetector, a general-purpose object-detection model trained to locate animals, people, and vehicles in camera-trap images regardless of species. Rather than classifying species, MegaDetector filters the enormous volume of empty frames, often the majority of captures triggered by wind or moving vegetation, and crops detections for downstream review or classification. This animal-versus-empty filtering dramatically reduces the manual effort of processing camera-trap data and generalizes far better across new locations than species classifiers, because detecting that something is present is easier than naming it. MegaDetector became one of the most widely adopted tools in the field. The paper is a key reference underpinning many practical camera-trap AI pipelines.