"Camera traps across Africa generate millions of images, but up to 75% are false triggers, and the species most critical to conservation. Endangered, nocturnal, and rare, are the ones current AI models fail to detect. Tools like SpeciesNet perform well on common animals yet miss the majority of leopards, pangolins, and other vulnerable species. These mistakes are buried in global accuracy metrics but profoundly distort population monitoring.

Our project addresses this by integrating IUCN threat metadata, strengthening representations of rare species through taxonomic and few-shot fine-tuning, and building uncertainty-aware models that flag doubtful cases for ranger review, improving reliability and field impact."