Survey Examines Abstract Concept Classification in Computer Vision Research

A detailed survey paper has been released on arXiv (arXiv:2308.10562v2) that meticulously evaluates research concerning high-level visual comprehension in computer vision, emphasizing the role of abstract ideas in automatic image classification. As the domain of computer vision increasingly shifts towards tasks involving high-level visual sensemaking, the exact nature of these tasks is often unclear. This survey clarifies that uncertainty through a multidisciplinary approach, organizing findings into categories such as commonsense, emotional, aesthetic, and inductive interpretative semantics. By identifying and classifying tasks linked to high-level visual sensemaking, the paper sheds light on various research areas and explores how abstract concepts like values and ideologies are addressed, highlighting the challenges present in this evolving field.