Weakly Supervised Learning for Object Recognition: Investigating weakly supervised learning techniques for object recognition tasks using only partial or noisy annotations

Abstract

Weakly supervised learning has emerged as a promising approach for object recognition tasks, particularly when full supervision with precise annotations is challenging or impractical. This paper explores various weakly supervised learning techniques tailored for object recognition using only partial or noisy annotations. We discuss the motivation behind weakly supervised learning, its challenges, and review recent advancements in the field. Additionally, we provide a comparative analysis of different approaches, highlighting their strengths and weaknesses. Through empirical evaluations on benchmark datasets, we demonstrate the effectiveness of these techniques in achieving competitive performance compared to fully supervised methods. Our findings suggest that weakly supervised learning holds great potential for improving object recognition systems, especially in scenarios where obtaining high-quality annotations is difficult.