Few-Shot Object Recognition

A machine learning approach in which a model learns to identify a novel object from only a handful of labelled examples (commonly one to ten) rather than the hundreds or thousands typical of conventional supervised training. Few-shot object recognition underpins teachable and personalized recognition systems used by blind and low-vision people, because end users cannot realistically capture large training sets of every personal item they want their phone to find. Datasets like ORBIT and ORBIT-India are designed specifically to benchmark few-shot performance under realistic, user-captured conditions.