Márcio Oliveira

A research team in Tanzania had a backlog of 11 million wildlife photos. An open-source AI model processed the entire decades-long archive in a few days. Biologists rely on motion-triggered cameras to see what animals do when humans aren't around. The cameras capture incredible footage. They also create a massive logistical burden. A single project easily generates millions of snapshots. For years, turning those images into usable data meant paying experts or begging volunteers to manually tag every single frame. Google open-sourced SpeciesNet a year ago. The AI model automatically identifies nearly 2,500 categories of mammals, birds and reptiles. The Snapshot Serengeti project had operated camera traps since 2010. They tried using online volunteers, but the sheer volume of images overwhelmed them. Researchers ran SpeciesNet on their 11 million backlogged photos. The model processed decades of data in days, finally giving the team a comprehensive, long-term view of animal populations in the region. The Humboldt Institute in Colombia analysed their own massive image datasets through the tool. They discovered rapid, undocumented shifts in animal behaviour. Mammals in the Amazon are becoming nocturnal to avoid threats. Birds in developed areas are delaying their morning routines to avoid predators. In Australia, the Wildlife Observatory took the open-source code and adapted it. They trained the model to identify endemic species like cassowaries and pademelons. Local conservationists can take the base architecture and fine-tune it for the specific endangered animals in their own backyards. State agencies like the Idaho Department of Fish and Game use the model to sort millions of images annually. The AI does the heavy lifting, and human experts handle the final review.