Deep Visual Recognition with Limited Human Supervision

This thesis explores how to make computer vision models, which help computers understand visual data, more effective when we have only a small amount of human annotated data to teach them. It investigates various strategies to improve the way these models learn from limited examples, presenting novel approaches that make the learning process more efficient and accurate. These innovations hold significant potential to advance computer vision technology in practical, real-world applications where labeled data can be scarce or expensive to obtain.