PoseSwin: Revolutionizing Bear Tracking with Deep Learning

0:00Bear biologist Beth Rosenberg just dropped a game-changer for studying Alaska's brown bears. She created Postwin, a deep learning system that IDs individual bears using only photos, no collars, tags, or samples needed. Trained on 73,000 images from McNeil River Sanctuary, it spots 109 unique bears across seasons and years. Rosenberg built the massive photo database over years at the Sanctuary, teaming up with her advisor Nathan Wolfe and Swiss computer experts from Echoe. Polytechnic, Federerau, Delozon. They published their findings in current biology this February of 2026, proving photo ID works for unpatterned animals like bears. Staff and visitors at places like McNeil have long recognized regular bears by sight, despite fur changes or weather shifts. This tech takes test skills statewide, focusing on stable head features like muzzle shape, brow bone angle, ear placement, and scars. Test show Postwin not only re-id's known bears, but spots new ones too,

1:06even tracking movements nearly 50 miles between sites like McNeil and Katmai using tourist photos. Citizen science could supercharge it with public uploads. This hands-off tool adds firepower to bear research, boosting conservation by revealing movements, behaviors, and population insights without, stressing the animals paving the way for smarter management everywhere bears roam. That's your update from Anchorage News Today, powered by AI.