Abstract by David Kartchner
Deep Object Localization Transfer Learning for Automated Wildlife Monitoring
Transfer learning has shown to meaningfully improve performance in multiple domains, including natural language processing, image recognition, and automated disease diagnosis. We demonstrate the power of transfer learning from widely-used object localization datasets such as PASCAL 2012 and MS COCO on data collected from remote, automated wildlife monitoring stations operated by the United States Air Force. We show that networks can quickly transfer knowledge learned from standard object detection tasks to quickly learn, recognize, and count never-before-seen objects.